1. A unique, explicit identification of a memory location, a peripheral device, or a location within a peripheral device. 2. A precise memory location that is an actual address number rather than an expression from which the address can be calculated.

One of two modes of execution for an application on UNICOS/mp systems; the other mode is flexible mode. Applications running in accelerated mode perform in a predictable period of processor time, though their wall clock time may vary depending on I/O usage, network use, and/or whether any oversubscription occurs on the relevant nodes. Due to the characteristics of the memory address space, accelerated applications must run on logically contiguous nodes. See also flexible mode.

The fabric components in a present topology that the administrator accepts for logical topology planning.

An attribute of a node, mostly controlled by administrators, that determines whether end users can log in or submit jobs to the node, depending on the partition’s attributes. The access control attribute of a partition also affects end-user access.

An attribute of a partition, controlled by administrators, that determines whether end users can log in or submit jobs to the partition, depending on the partition’s characteristics. The access control attribute of a node also affects end-user access.

An extension to the normal file discretionary access control. ACLs support the ability to grant or deny access to a file on any user and/or group basis.

Configuration information that the Active Manager software maintains to control which Linux groups can access the Cray XD1 system. Groups are granted access either to the entire system (administrator privileges) or to one or more partitions.

A user action on an alarm to indicate that the user is aware of the alarm. The user can choose to filter acknowledged alarms out of an alarm list.

In the Active Manager GUI, a simple function that a user performs by clicking a button in the workspace of a page.

A point in a program at which a breakpoint, evaluation point, or event point has been set during a debugger session.

The state of an alarm when the condition that triggered it still exists.

The software that monitors and manages all aspects of the Cray XD1 system. Its user interfaces provide administrators and end users with a single point of control for the system.

The relational database in which the Active Manager software stores information necessary to manage the Cray XD1 system.

A component of the Active Manager architecture; a J2EE-based application server that runs on one or more nodes. The Active Manager server implements the Active Manager business logic.

A Cray XD1 node that is allocated to a partition and is currently booted.

In SHMEM, a set that is defined to participate in a collective operation.

A user of the Cray XD1 system with unlimited access privileges, including permission to issue all Active Manager commands. The administrator is responsible for monitoring and managing the system.

The browser-based Active Manager GUI for Cray XD1 administrators. It presents the full set of GUI capabilities, in contrast with the User Portal.

A type of component of the Active Manager architecture; a lightweight remote worker that communicates with the Active Manager server to monitor or control one aspect of the system. Some agents run on every node in the Cray XD1 system.

An Active Manager process that launches agents as necessary on a node.

For Cray X1 series systems, the description of the cabinet that uses a combination of Fluorinert liquid and forced air to cool the components on a compute module. The Fluorinert is air cooled within a heat exchange unit so that a chilled water supply is not required for AC Cray X1 series systems. An AC cabinet uses two blowers: one for the forced air cooling in the module cabinet, and the second for the Fluorinert heat exchange unit.

A record that the Cray XD1 system generates when a monitored attribute meets the condition specified in an alarm definition.

An Active Manager object that specifies the condition on a monitored attribute that generates an alarm. The Cray XD1 system has a default set of alarm definitions, and an administrator can create additional definitions.

A Cray XD1 node that is assigned to a partition. Its working software image is based on the master software image for the partition.

A software package in the Cray XD1 software distribution that includes 3 mathematical libraries: Basic Linear Algebra Subprograms (BLAS), Fast Fourier Transform (FFT), and Linear Algebra Package (LAPACK). Routines are available for both Fortran 77 and C interfaces.

See FPGA application acceleration processor.

The components of the Active Manager software that run on nodes: the Active Manager server and associated components.

For UNICOS/mp systems, a node that is used to run user applications. Application nodes are best suited for executing parallel applications and are managed by the strong application placement scheduling and gang scheduling mechanism Psched. See also node; node flavor.

The major unit of distribution of software for the Cray XD1 system. It contains HPC-optimized Linux, the Active Manager software, the FPGA logic design framework and sample FPGA logic, and any third-party applications bundled by Cray. The application release package consists of RPM packages.

Applications launched via the aprun or mpirun command create an ApTeam of processes. An ApTeam can be a single execution thread or a large number of processes and can use shared memory, distributed memory, or a hybrid model for parallel execution.

See array syntax statement.

A Fortran statement that allows you to use the array name (or the array name with a section subscript) to specify actions on all the elements of an array (or array section) without using DO loops. For example, the A = B array syntax statement assigns all the values of array A to array B. Sometimes called an array assignment statement.

The set of information used in Fortran to alter the details of a Fortran connection. This information includes a list of unit numbers, file names, and file name patterns that have attributes associated with them. Any file name, file name pattern, or unit number to which assign options are attached is called an assign object. When the unit or file is opened from Fortran, the options are used to set up the properties of the connection.

In Fortran, a dummy or pointer array whose last dimension is of an unknown size (specified as *) that is assumed to be large enough for all references made to the array.

I/O operation during which the program performs other operations that do not involve the data in the I/O operation. Control is returned to the calling program after the I/O is initiated. The program may perform calculations unrelated to the previous I/O request, or it may issue another unrelated I/O request while waiting for the first I/O request to complete. An operation is complete when all data has been moved.

Making a remote file system accessible (using NFS) when the file system is accessed.

A node that is physically capable of becoming the master node and is configured to store a periodically synchronized copy of the Active Manager data stores. The administrator can quickly put such a node into service as the master node if the current master node fails or is taken out of service. The enrollment of any node as a backup node is optional; the administrator can enroll one or more suitable nodes, or none.

An obstacle within a program that provides a mechanism for synchronizing tasks. When a task encounters a barrier, it must wait until all specified tasks reach the barrier.

In code, a barrier is used after a phase. The barrier delays the streams that were executing parallel operations in the phase until all the streams from the phase reach the barrier. Once all the streams reach the barrier, the streams begin work on the next phase.

1. An event initiated by software that prevents cooperating tasks from continuing to issue new program instructions until all of the tasks have reached the same point in the program. 2. A feature that uses a barrier to synchronize the processors within a partition. All processors must reach the barrier before they can continue the program.

The starting absolute address of the memory field that is assigned to the user's job. This address is maintained in the base address register. A base address is loaded into a register and used as a starting point from which instructions form addresses. The operating system assigns the absolute starting address to a program. In Cray systems, there are two registers: a database address register, which is used as the starting point for data memory references, and an instruction base address register, which is used as the starting point for instruction fetches.

A section of a program that does not cross any conditional branches, loop boundaries, or other transfers of control. There is a single entry point and a single exit point. Many compiler optimizations occur within basic blocks.

A version of the UNIX operating system developed at the University of California at Berkeley; for example, 4.2BSD and 4.3BSD.

A file format that describes blocked, nontranslatable data.

An ordered sequence of characters that can transparently record internal data. Data read in from a binary stream equals data that was written earlier out to that stream under the same implementation.

The way in which one component in a resource specification is related to another component.

1) A field-replaceable physical entity. A Cray X2 compute blade consists of eight Cray X2 chips (CPU and network access links), two voltage regulator modules (VRM) per CPU, 32 memory daughter cards, a blade controller for supervision, and a back panel connector. 2) From a system management perspective, a logical grouping of nodes and blade control processor that monitors the nodes on that blade.

1) A field-replaceable physical entity. A Cray XMT service blade consists of AMD Opteron sockets, memory, Cray SeaStar chips, PCI-X or PCIe cards, and a blade control processor. A Cray XMT compute blade consists of Threadstorm processors, memory, Cray SeaStar chips, and a blade control processor. 2) From a system management perspective, a logical grouping of nodes and blade control processor that monitors the nodes on that blade.

1) A field-replaceable physical entity. A service blade consists of AMD Opteron sockets, memory, Cray network application-specific integrated circuit (ASIC) chips, PCI cards, and a blade control processor. A compute blade consists of AMD Opteron sockets, memory, Cray network application-specific integrated circuit (ASIC) chips, and a blade control processor. 2) From a system management perspective, a logical grouping of nodes and blade control processor that monitors the nodes on that blade.

A microprocessor on a blade that communicates with a cabinet control processor through the HSS network to monitor and control the nodes on the blade. See also blade, L0 controller, Hardware Supervisory System (HSS).

A type of Fortran program unit. A block data program unit contains only data definitions. It specifies initial values for a restricted set of data objects.

Method of thread execution used by the compiler where contiguous blocks of loop iterations are divided equally and assigned to available streams. For example, if there are 100 loop iterations and 10 streams, the compiler assigns 10 iterations to each stream. The advantage to this method is that data in registers can be reused across adjacent iterations rather than releasing a stream after each iteration.

A special file that provides an interface to a device that is capable of supporting a file system.

A command that waits for the requested function to complete before it terminates. Most Active Manager commands are blocking, but a few that perform lengthy functions offer a nonblocking option.

An optimization that involves changing the iteration order of loops that access large arrays so that groups of array elements are processed as many times as possible while they reside in cache.

The node from which the UNICOS/mp operating system is or can be booted; typically an OS node. See also node; OS node; system node.

A unit of distribution of embedded software for the Cray XD1 system. It contains firmware components that are installed only at the factory and one field-installable component called DAMON.

A point in a program that, when reached, triggers some special behavior useful to the process of debugging; generally, breakpoints are used to either pause program execution and/or dump the values of some or all of the program variables. Breakpoints may be part of the program itself, or they may be set by the programmer as part of an interactive session with a debugging tool for scrutinizing the execution of the program.

A Cray X1 series hardware term for four grouped compute modules. On the Cray X1 series mainframe cabinet, a brick is designated as either V or W. See also brick-pair; compute module.

For Cray X1 series systems, a component of a brick in the mainframe that includes the Fluorinert pump, filters, the reservoir/heat exchanger unit, supply and return manifolds, and a Cray L1 controller.

For Cray X1 series systems, two bricks (the space required for 8 compute modules) in the front or back half (V-side or W-side for notational reference) of a liquid-cooled Cray X1 series mainframe cabinet. Air-cooled Cray X1 series systems have only the V-side brick. See also brick; compute module.

For Cray X1 series and Cray X2 systems, the part of the C and C++ Programming Environment that translates C and C++ programs, respectively, into object files. The cc (C compiler), CC (C++ compiler), c89 (POSIX 1003.2-1992), and cpp (preprocesssor only) commands work with the C and C++ compilers.

A Fortran feature that allows Fortran programs to call C functions and access C global objects and also allows C programs to call Fortran procedures and access Fortran global objects.

A Fortran feature that allows Fortran programs to call C functions and access C global objects and also allows C programs to call Fortran procedures and access Fortran global objects.

For Cray X1 series systems, the modular controller component of a RAID subsystem, also called a controller brick. It is located in a PC-20. A C-brick can have multiple associated RAID storage components (S-bricks).

A microprocessor in the cabinet that communicates with the HSS via the HSS network to monitor and control the devices in a system cabinet. See also Hardware Supervisory System (HSS).

The subset of memory that is permitted to be cached. For Cray X1 systems, all the memory on the node where the multistreaming processor (MSP) resides. For Cray X1E systems, the cache domain for an "upper" MSP is the upper half of memory on its node, and the cache domain for a "lower" MSP is the lower half of memory on its node.

A division of cache. Each cache line can hold multiple data items. For Cray X1 series systems, a cache line is 32 bytes, which is the maximum size of a hardware message.

A division of cache. Each cache line can hold multiple data items. For Cray X1 systems, a cache line is 32 bytes, which is the maximum size of a hardware message.

A delay that results from loading data into cache that will not be used before it is evicted, thereby displacing cached data that would have been used.

A chassis in a cabinet of a Cray system. See chassis.

A chassis on a Cray XMT series system. See chassis.

A software stack of functions and subroutines used by the executing program. The functions and subroutines are listed in the opposite order in which they were called. That is, the function at the bottom of the stack is the one currently executing. When function a immediately follows function b in the list, a was called by b.

The operating system kernel developed by Sandia National Laboratories and implemented to run on Cray XT single-core compute nodes. See also Catamount Virtual Node (CVN); compute node.

The Catamount kernel enhanced to run on dual-core Cray XT compute nodes.

Stores current system state information from the event handler. The HSS uses this information to display the system state on the System Management Workstation (SMW). See also Hardware Supervisory System (HSS).

See processor.

A special file that provides an interface to an I/O device. The character interface is used for devices that cannot support a file system and as an alternative interface to devices capable of supporting a file system.

A blade enclosure together with power distribution and four R1 router cards. A chassis houses from one to eight blades (either compute or bridge blades). Cabinets contain chassis or router modules: a compute cabinet contains two chassis; a router cabinet has three chassis and contains a router backplane housing up to 16 router modules.

The hardware component of a Cray cabinet that houses blades. Each cabinet contains three vertically stacked chassis, and each chassis contains eight vertically mounted blades. See also cage.

The hardware component of a Cray XMT cabinet that houses blades. Each cabinet contains three vertically stacked chassis, and each chassis contains eight vertically mounted blades. See also cage.

The physical base unit of a Cray XD1 system. Includes a main board, a maximum of 31 commodity and specialized processors, one or two high-speed switch fabrics, a power supply, and four fans. May also contain a maximum of six SATA disk drives and a maximum of four PCI-X cards. Each Cray XD1 chassis occupies 3 vertical units in a 23-inch cabinet.

The permanent numeric identifier of a chassis, unique to each Cray XD1 chassis. A chassis ID has a maximum of six decimal digits.

An external RapidArray switch, required for Cray XD1 systems that have a fat-tree topology.

A group of service nodes of a particular type, such as login or I/O. See also specialization.

The operating system for Cray XE and Cray XT systems.

A user action on an alarm to indicate that the condition that triggered the alarm no longer exists. Applies only to certain user-defined alarms that the Active Manager software cannot automatically detect as cleared.

The state of an alarm when the Active Manager software or a user action indicates that the condition that triggered the alarm no longer exists.

An alarm record that matches a previously generated alarm to indicate that the previous alarm is cleared.

For a partition: The setting of the access control attribute of a partition that does not permit end users to log in or submit jobs, regardless of the partition's characteristics.
For a node: The setting of the access control attribute of a node that does not permit end users to log in or to submit jobs to the node, regardless of the characteristics of its partition.

The CLE compute node kernel. CNL provides a set of supported system calls. CNL provides many of the operating system functions available through the service nodes, although some functionality has been removed to improve performance and reduce memory usage by the system.

A syntactic extension to Fortran that offers a method for programming data passing; a data object that is identically allocated on each image and can be directly referenced syntactically by any other image.

The dimensions of a co-array; specified within brackets ([ ]). A co-array specification consists of the local object specification and the co-dimensions specification.

An Active Manager agent that processes telemetry data from multiple monitoring agents.

The set of Active Manager commands that a user can issue from the Linux command line to manage and monitor the Cray XD1 system.

The process by which an administrator or installer prepares a Cray XD1 system or component for use after it is physically installed.

An area of memory, or block, that can be referenced by any program unit. In Fortran, a named common block has a name specified in a Fortran COMMON or TASKCOMMON statement, along with specified names of variables or arrays stored in the block. A blank common block, sometimes referred to as blank common, is declared in the same way but without a name.

A graphical user interface (GUI) for loading software. The cit and setup commands invoke CIT.

One of six circuit boards in a Cray XD1 chassis; contains Opteron processors configured as an SMP, DIMMs, and a RapidArray processor. A compute blade may also have an expansion module.

See blade.

For a Cray X1 series mainframe, the physical, configurable, scalable building block. Each compute module contains either one node with 4 MCMs/4MSPs (Cray X1 modules) or two nodes with 4 MCMs/8MSPs (Cray X1E modules). Sometimes referred to as a node module. See also node.

A node that runs application programs. A compute node performs only computation; system services cannot run on compute nodes. Compute nodes run a specified kernel to support either scalar or vector applications. See also node; service node.

The runtime environment available to compute nodes for use in dynamic compiling, linking and execution of programs.

A Data Virtualization Services (DVS) server that projects the shared root to compute nodes for use with dynamic shared objects.

A partition with a disabled login attribute and an enabled job execution attribute. Such a partition is suitable for running computing jobs.

See processing element.

Arrays that have the same rank (number of dimensions) and dimension extents and that have the same assigned weight to each dimension.

A variable that is incremented only by a loop-invariant value. For example, in a loop with index J, the statement J = J + K, in which K can be equal to 0, J is a CIV.

A sequence of statements in Fortran that starts with a SELECT CASE, DO, IF, or WHERE statement and ends with the corresponding terminal statement.

An Active Manager agent that issues commands to one component of the system.

An action statement (such as a GO TO, STOP, or RETURN statement or a CASE, DO, or IF construct) that can change the normal execution sequence.

The component of the system interconnection network that provides message routing and communication services. See also system interconnection network.

See Cray application-specific integrated circuit

The software program that assembles Cray Assembly Language files into Cray X1 or Cray X2 object files. (The Cray X1 and Cray X2 Assembly Language and the Cray X1 and Cray X2 assembler are both referred to as CAL.) Unlike previous versions of CAL, the Cray X1 and Cray X2 assembler allows for explicit definition of data types as 32-bit or 64-bit and as integer or floating point and uses hexadecimal rather than octal for addresses and displayed machine code. The assembler recognizes a set of pseudoinstructions, macros, micros, and opdef commands and directives.

The Cray Data Virtualization Service (Cray DVS) is a distributed network service that provides compute nodes with transparent access to file systems on the service partition using the Cray high-speed network.

The compiler that translates Fortran source code into executable object files. The Cray Fortran Compiler supports the Fortran language through the Fortran 2003 Standard, ISO/IEC 1539-1:2004, with some exceptions and extensions, as well as selected features of the proposed Fortran 2008 Standard, as described in the Cray Fortran user documentation.

For Cray X1 series systems, a microprocessor, physically located on a Cray L1 controller assembly and connected to a Cray L1 controller, that provides environmental monitoring and protection for the compute module, router module, brick cooling unit (BCU), and I/O drawer (IOD) components.

For Cray X1 series systems, a microprocessor, embedded on a Cray L1 controller assembly, that is used as the maintenance access point to the compute module, router module, brick cooling unit (BCU), and I/O drawer (IOD) components.

For Cray X1 series systems, a hardware assembly that contains both the Cray L1 controller and the Cray L0 controller, which are used to control power, initialize, diagnose, and monitor the state of a module and its surrounding environment. Each compute module, router module, brick cooling unit (BCU), and I/O drawer (IOD) has a Cray L1 controller assembly.

A specialized router (gateway) providing IP (Internet Protocol) network connectivity between a supported Cray mainframe and site networks. The CNS also provides Transmission Control Protocol (TCP) assist functionality that can enhance TCP performance in supported Cray systems.

A collection of public-domain software that is an optional product available for certain Cray systems. COS provides options that are either not available in or behave differently than Cray operating system utilities.

See Cray pointer.

A variable whose value is the address of another entity, which is called a pointee. The Cray pointer type statement declares both the pointer and its pointee. The Cray pointee does not have an address until the value of the Cray pointer is defined; the pointee is stored starting at the location specified by the pointer.

A server for the Cray X1 series system that runs the Programming Environment software.

Deprecated term: Software and hardware that implements reliability, availability, and serviceability (RAS) and some system management functions. The CRMS components use a private Ethernet network, not the system interconnection network. See also system interconnection network, Hardware Supervisory System (HSS).

The component of the system interconnection network that provides message routing and communication services. See also system interconnection network.

Software on the Cray Workstation (CWS) that is used for configuring the RAID subsystem.

Nonadvisory directives that allow you to more closely control multistreaming for key loops.

For Cray X1 series systems, the system operation, administration, and maintenance workstation.

The Cray system that combines the single-processor performance and single-shared address space of Cray parallel vector processor (PVP) systems with the highly scalable microprocessor-based architecture that is used in Cray T3E systems. Cray X1 and Cray X1E systems utilize powerful vector processors, shared memory, and a modernized vector instruction set in a highly scalable configuration that provides the computational power required for advanced scientific and engineering applications.

A Cray system that uses Cray X2 compute nodes for user application processing and Cray XT series service nodes for login, network, I/O, and boot functions.

A stand-alone Cray XD1 chassis or multiple chassis that communicate over both the supervisory network and the RapidArray interconnect.

A 64-bit router chip that provides routing and communication services between Cray X2 compute nodes. The YARC interconnect enables systems to be scaled using a fat-tree topology.

Cray's documentation system for accessing and searching Cray books, man pages, and glossary terms from a web browser.

For Cray X1 series systems, the primary high-level tool for identifying opportunities for optimization. CrayPat allows you to perform profiling, sampling, and tracing experiments on an instrumented application and to analyze the results of those experiments; no recompilation is needed to produce the instrumented program. In addition, the CrayPat tool provides access to all hardware performance counters.

An online technical-assistance and problem-reporting service for subscribing Cray customers.

An alarm that indicates a fault condition that requires immediate corrective action. A component may be completely out of service or is at immediate risk of going out of service.

A communications switch that provides direct connection between any pair of ports.

A Cray XD1 node that is not managed by the Active Manager software. A custom node resides in a custom partition.

A partition that contains nodes that are not managed by the Active Manager software.

The Cray X1 series system internal Ethernet network that connects the Cray Workstation to the access point to be used by Cray Customer Support for remote access to your Cray Workstation. It requires an Ethernet connection for an ISDN router.

The network that connects Cray Workstations (CWSs) together through a switch. This CWS shared private network exists only when there are multiple CWSs connected to a single Cray X1 series system, due to either partitioning or resiliency.

For Cray X1 series systems, a small, high-speed random-access memory that stores frequently or recently accessed data (data cache). The second level of cache storage for scalar operations, it is located between the registers and the E-cache.

An embedded software component that controls part of the chassis startup process and also provides a "safe-mode" interface for technical support personnel.

Directly attached storage.

Transferring data from one object to another; useful for programming single-program-multiple-data (SPMD) parallel computation. Its chief advantage over message passing is lower latency for data transfers, which leads to better scalability of parallel applications. Data passing can be achieved by using SHMEM library routines or by using co-arrays.

The process by which an administrator or installer prepares a Cray XD1 system for the physical removal of a component or the complete dismantling of the system. This process puts the system or component into a state from which it can be recommissioned, but does not necessarily restore the factory state.

The label used to introduce information about a feature that will not be implemented until a later release.

A technique used by the compiler to determine if any iteration of a loop depends on any other iteration (this is known as a loop-carried dependency).

The term deprecated is applied to software features or functionality that will be removed in a future release and should be avoided. Although deprecated features remain in the current release, alternate or new features are usually recommended to replace the functionality. Cray books or man pages document recommended alternatives. Features are deprecated before being removed in order to provide backward compatibility and give users time to convert to the recommended replacement.

The name that an administrator assigns to a Cray XD1 chassis. The default designation is the chassis ID. The designation of a chassis usually displays on the second line of its LCD.

Also known as "all-to-all mesh." A switchless multichassis topology in which every Cray XD1 chassis in the system connects to every other chassis. A maximum of 13 chassis can be directly connected in a Cray XD1 system.

A feature of the Cray XD1 system in which the RapidArray interconnect provides high-speed interconnections among all processors in the system, which eliminates the need for slower I/O connection paths.

A technique that allows a peripheral device or a channel to control data transfer in and out of the computer. The data transfers directly to or from storage and bypasses the processor.

I/O operation where the a peripheral device or a channel controls data transfer in and out of the computer. The data transfers directly to or from storage and bypasses the processor.

A circuit board of a Cray XD1 chassis that holds one or two SATA disk drives.

Provides connectors for three disk blades in a Cray XD1 chassis. The disk expansion card is in chassis that do not have the PCI-X expansion card.

An optimization technique that allows an applications program to overlap computation and I/O. The I/O libraries set up two data buffers for I/O requests. During a read, for example, one buffer will be used for I/O while computations take place on data from the other buffer.

Four bytes (32 bits).

A processor that combines two independent execution engines ("cores"), each with its own cache and cache controller, on a single chip.

In OpenMP, an extent that includes both the statements of a lexical extent and the statements of a function that is called from the lexical extent. A dynamic extent is an instance of a region.

In a dynamic schedule, the compiler does not bind iterations to streams at loop startup. Instead, streams compete for each iteration using a shared counter.

See dynamic shared object.

A program resource that can be linked and loaded into the address space of one or more applications at runtime.

In OpenMP, the automatic adjustment of the number of threads between parallel regions. Also known as dynamic threads or the dynamic thread mechanism.

For Cray X1E systems, a local, external cache for the 4 P+ chips in a multichip module (MCM). Each E+ chip contains an upper half and a lower half. Each half communicates with the corresponding half of the P+ chips and the corresponding half of memory. The external cache consists of 4 E+ chips that are mapped to separate quarters of the node's memory. Each E+ chip maintains a working subset of the most recently accessed memory on the node board, resulting in a reduced memory read latency on cached data and a reduced dependency on system memory bandwidth for applications with a high cache hit rate. See also E-cache; E-chip.

For Cray X1 series systems, the first level of cache storage. It is located between node memory and a multistreaming processor (MSP). It is provided by the E-chips (for Cray X1 systems) or E+ chips (for Cray X1E systems) in a multichip module and shared by all processors in an MSP. All data that enters an MSP routes through the E-cache. Scalar, vector, and instruction data can be stored in the E-cache. For Cray X1E systems, the E-cache maintains the same functionality as the X1 E-cache but is organized differently. The E-cache is organized into two logical halves: upper and lower. The upper half caches data from the upper half of module memory and the lower half caches data from the lower half of memory. The 4 upper halves of the P+ chips use the upper half of the cache and the 4 lower halves of the P+ chips use the lower half of the cache. This means that data in the upper half of memory accessed by a P+ lower half is accessed uncached and vice-versa.

For Cray X1 systems, a local, external cache for the four P-chips in a multichip module (MCM). The external cache consists of four E-chips that are mapped to separate quarters of the node's memory. Each E-chip maintains a working subset of the most recently accessed memory on the compute module, resulting in a reduced memory read latency on cached data and a reduced dependency on system memory bandwidth for applications with a high cache hit rate. See also E-cache; E+ chip.

I/O statements where the format provides information that directs the editing between internal representation and the character strings of a record (or sequence of records) in the file.

The amount of real time, as measured by a clock, that passes between the start and termination of a user process. This includes user processing time, operating system time, I/O wait time, and sleep or idle time.

A user of the Cray XD1 system who does not have administrator privileges.

A location in a program or routine at which execution begins. A routine may have several entry points, each serving a different purpose. Linkage between program modules is performed when the linkage editor binds the external references of one group of modules to the entry points of another module.

A variable that stores a string of characters for use by your shell and the processes that execute under the shell. Some environment variables are predefined by the shell, and others are defined by an application or user. Shell-level environment variables let you specify the search path that the shell uses to locate executable files, the shell prompt, and many other characteristics of the operation of your shell. Most environment variables are described in the ENVIRONMENT VARIABLES section of the man page for the affected command.

Optional Cray XD1 hardware that connects to each compute blade; if they are present, a chassis has six expansion modules. The expansion modules provide a node with a second RapidArray processor, two additional Rapid Array links, and an optional application acceleration processor.

An XML file that specifies the intended physical topology of the RapidArray interconnect. The Active Manager software uses an expected topology file to verify the present or accepted topology. The file lists all the Cray XD1 chassis, switch chassis, and external RapidArray links in the system. An expected topology file may be a template, which uses logical IDs to identify chassis, or a physical topology file that uses hardware IDs to identify specific chassis. The system includes expected topology templates for standard system configurations.

In CrayPat, a predefined entity that specifies what performance data to collect.

One classification of CrayPat experiments that defines how an instrumented program collects data. The types are trace and sampling.

A programming style in which communication is language independent and in which communication occurs through library calls. The message-passing and explicit shared memory programming methods use products such as MPI or SHMEM.

The process by which the user performs calls to subroutines that convert native data to and from foreign data formats.

The opening of a file or a unit when the first reference to a unit number is an OPEN statement without a FILE specifier.

The collection of fabric components that interconnect in the same switching plane. A Cray XD1 system has one or two independently wired, parallel RapidArray fabrics: the main fabric and the optional expansion fabric. These fabrics are also known as fabric X and fabric Y, respectively.

A Cray XD1 hardware entity that is managed by the fabric manager. Fabric components include RapidArray switches, Cray XD1 chassis, switch chassis, RapidArray processors, RapidArray ports, and RapidArray links.

Optional hardware in a Cray XD1 chassis that adds a second RapidArray fabric to the system: provides a second internal 24-port RapidArray switch, 12 additional internal links, and 12 additional external ports for chassis interconnection. The fabric expansion card connects to the main board.

An Active Manager component that manages and monitors the Cray XD1 RapidArray fabric components. There is a single fabric manager in a Cray XD1 system, regardless of whether the system has one or both fabrics.

See path, fabric.

The ability to define and manage alternate paths to a single disk device or part of a disk device.

One of two perforated covers on the front of a Cray XD1 chassis. Both are hinged and can be flipped down to provide access to the fans. The right-side fascia also holds the LCD.

A physical topology that uses a hierarchy of switches to limit latency and achieve good bisection bandwidth in the system. Cray XD1 systems that exceed 13 chassis are configured as a fat tree.

The set of actions that the system performs when an alarm is generated that has a fault response definition linked to it. Also refers to the Active Manager object that records the actions.

An Active Manager object that specifies the actions that the system takes when an alarm of a specific type is generated. The Cray XD1 system has a default set of fault response definitions, and an administrator can create additional definitions.

Optional hardware in a Cray XD1 chassis that plugs into a PCI-X slot to provide a Fibre Channel interface for a device such as a SAN.

A Cray document that communicates technical or procedural information to Cray customers, Cray employees, and third-party service providers about Cray products.

A document that accompanies a field-replaceable unit to provide the instructions for installing the unit.

An integrated circuit that consists of arrays of AND and OR gates (typically thousands) that can be programmed to perform complex functions. The Cray XD1 system has optional FPGAs available for use as application acceleration processors.

A component of the system that a customer can replace at the customer site.

A method of I/O, sometimes called layered I/O, wherein each processing step requests one I/O layer or grouping of layers. A layer refers to the specific type of processing being done. In some cases, the name corresponds directly to the name of one layer. In other cases, however, specifying one layer invokes the routines used to pass the data through multiple layers.

One of two modes of execution for an application on UNICOS/mp systems; the other mode is accelerated mode. Applications running in flexible mode may run on noncontiguous nodes; they perform in a less predictable amount of processor time than applications running in accelerated mode due to the exclusive use of source processor address translation. See also accelerated mode.

A basic compiler optimization that converts operations on constants to simpler forms as these examples show: Operation to fold Folded operation 1 + 2 3 5.0/3.0 + 1.7 3.366... (if the -O fp1 (Fortran) or -h fp1 (C/C++) or greater is used.) sin( 1.3 ) 0.96355818... 3 + n - 4 n - 1

An option on some Active Manager commands; specifies that the command must execute even though it may have disruptive results.

Occurs when processors allocate additional streams to a thread at the point where it is creating new threads for a parallel loop operation.

Data transfer with editing. Formatted I/O can be edit-directed, list-directed, or namelist I/O. If the format identifier is an asterisk, the I/O statement is a list-directed I/O statement. All other format identifiers indicate edit-directed I/O. Formatted I/O should be avoided when I/O performance is important.

A DNS server to which a local server forwards requests to resolve names in specified domains. The recommended configuration in the Cray XD1 system is to configure a server on the site network as a forwarder for all domains outside the Cray XD1 subdomain.

See logical topology.

An FPGA that users can program to accelerate computationally intensive and repetitive algorithms; acts as a co-processor to the Opteron processor. This is an optional component on the expansion module.

State of a variable that indicates whether it contains a value (full) or not (empty). Generic read and write operations use this state to determine whether they can perform an operation on the variable. For example, for a writeef operation can only write a value to a variable if the state is empty. After the write operation, it sets the state to full.

Implements user-specified or explicit parallelism by starting new threads. A future is a sequence of code that can be executed by a newly created thread that is running concurrently with other threads in the program. Futures delay the execution of code if the code is using a value that is computed by a future, until the future completes. The thread that spawns the future uses parameters to pass information from the future to the waiting thread, which then executes. In a program, the term future is used as a type qualifier for a synchronization variable or as a keyword for a future statement.

An operation that copies data between remote and local memory or within local memory. A gather is any software operation that copies a set of data that is nonsequential in a remote (or local) processor, usually storing into a sequential (contiguous) area of local processor memory. A scatter copies data from a sequential, contiguous area of local processor memory) into nonsequential locations in a remote (or local) memory.

An operation that copies data between remote and local memory or within local memory. A gather is any software operation that copies a set of data that is nonsequential in a remote (or local) processor, usually storing into a sequential (contiguous) area of local processor memory. A scatter copies data from a sequential, contiguous area of local processor memory) into nonsequential locations in a remote (or local) memory.

Optional hardware in a Cray XD1 chassis that connects to a PCI-X slot to provide Ethernet interfaces for connecting the system to site networks.

The ID used by applications executed in accelerated mode on UNICOS/mp systems. It is used to help translate remote virtual addresses to physical addresses at the destination node. See also accelerated mode.

See implicit communication.

The UNICOS/mp function that manages and allocates resources, including processors and memory.

From The Free Software Foundation, a compiler that supports C, C++, Objective-C, Fortran, and Java code (see http://gcc.gnu.org/ ).

A security feature that divides the power of the superuser into discrete units of privilege called capabilities.

The software that runs on the management processor in each Cray XD1 chassis. It primarily monitors the hardware components of the system and proactively manages the health of the system. It communicates with nodes and with the management processors in other chassis over the supervisory network.

Hardware and software that monitors the hardware components of the system and proactively manages the health of the system. It communicates with nodes and with the management processors over the private Ethernet network. See also system interconnection network.

A signal sent at regular intervals by software to show that it is still active.

For Cray X1 series systems, a cache for the processor in a P-chip (for Cray X1 systems) or in a P+ chip (for Cray X1E systems) that holds instructions that will soon be executed.

For Cray X1 series systems, an interface for two I/O channels on the compute module. I-chips transfer data between the M-chips (memory system) on compute modules and I/O channel adapters (IOCAs) on the I/O bridges (IOBs) on to the I/O devices (front-end I/O or local I/O).

For Cray X1 series systems, a board within an I/O drawer of an I/O chassis that acts as an I/O bridge to convert System Port Channel (SPC) data from/to PCI-X format. (Also known as an IOCA module because it contains two I/O Channel Adapters.)

For Cray X1 series systems, a field programmable gate array, resident on an IOB, that converts proprietary System Port Channel transactions into standard PCI-X bus transactions.

For Cray X1 series systems, the hardware enclosure that contains two I/O boards (IOBs) and a Cray L1 controller assembly. Each I/O drawer supports System Port Channels (SPCs) to the Cray X1 series mainframe cabinet and Fibre Channel connections to I/O devices. (Also known as an I/O module.)

A programming style in which most communication is recognized implicitly by the compiler and run-time libraries from the Fortran source (including directives). The data-sharing and work-sharing programming methods use implicit communication. Also called global addressing or implicit shared memory.

The process by which you declare that a particular file contains foreign data and/or record blocking and then request that the run-time library perform appropriate transformations on the data to make it useful to the program at I/O time.

The opening of a file or a unit when the first reference to a unit number is an I/O statement other than OPEN, CLOSE, INQUIRE, BACKSPACE, ENDFILE, or REWIND.

A variable that is increased or decreased by a fixed amount on each iteration of a loop.

An inductive loop is one which contains no loop-carried dependencies and has the following characteristics: a single entrance at the top of the loop; controlled by an induction variable; and has a single exit that is controlled by comparing the induction variable against an invariant.

The system software components that run on the management processor and programmable components other than nodes: primarily the hardware supervisory subsystem.

The major unit of distribution of embedded software for the Cray XD1 system. It contains software and firmware components for the hardware supervisory subsystem and the compute blades.

The software image that the Active Manager software uses to temporarily boot a node and install the node’s working software image on its local disk.

The unit that retrieves instructions for scalar operations.

See RapidArray interconnect.

Method of executing loop iterations used by the compiler where contiguous iterations are assigned to distinct streams. For example, for a loop with 100 iterations and 10 streams, one stream performs iterations 1, 11, 21,... while another stream performs iterations 2, 12, 22, ..., and so on. This method is typically used for triangular loops because it reduces imbalances. One disadvantage to using this method is that there is loss of data reuse between loop iterations because the stream is released at the end of the iteration.

A rule, such as the ordering of an ordered list or heap, that applies throughout the life of a data structure or procedure. Each change to the data structure must maintain the correctness of the invariant.

A computing task that runs on one processor or multiple processors concurrently. The workload management (WLM) system assigns the requested resources and launches the job.

An attribute of a Cray XD1 partition that indicates whether the WLM system is enabled; inherited by all nodes that are allocated to the partition. If this attribute is enabled, end users can submit jobs to the partition and the WLM system manages the job lifecycle.

A logical group of nodes assigned to a job and allocated from either the interactive or the batch pool of nodes.

Occurs when threads that are forked for a parallel operation finish the operation. As threads finish and drop the streams they are running on, the streams join back together until there is a single stream running the thread.

Optional hardware that provides a communication path to the JTAG port of the FPGA application acceleration processor. This card connects to one of the high-speed I/O slots on the main board of a Cray XD1 chassis. With this card, you can connect a workstation that runs appropriate debugging tools to the FPGA.

A decimal multiple equal to 103 bytes (1,000 bytes).

A binary multiple equal to 2¹⁰ bytes (1,024 bytes).

Data representation (for example, single precision, double precision). The kind of a type is referred to as a kind parameter or kind type parameter of the type. The kind type parameter KIND indicates the decimal range for the integer type, the decimal precision and exponent range for the real and complex types, and the machine representation method for the character and logical types.

See blade control processor.

See blade control processor.

See cabinet control processor.

See flexible file I/O (FFIO).

The scope within which a C or C++ directive is on or off and is bounded by the opening curly brace just before the declaration of the directive and the corresponding closing curly brace. Only applicable executable statements within the lexical block are affected as indicated by the directive. The lexical block does not include the statements contained within a procedure that is called from the lexical block.
This example code shows the lexical block for the inline directive:

void Example(void)
{
    #pragma inline  // inline state is active
    ...
    {
       // inline state is
still on?
    }

    {
      #pragma noinline // inline state is now off
       ...
    }
     // inline state
is back on
    ...
}

The scope within which a C or C++ directive is on or off and is bounded by the opening curly brace just before the declaration of the directive and the corresponding closing curly brace. Only applicable executable statements within the lexical block are affected as indicated by the directive. The lexical block does not include the statements contained within a procedure that is called from the lexical block.
This example code shows the lexical block for the inline directive:

void Example(void)
{
    #pragma inline  // inline state is active
    ...
    {
       // inline state is
still on?
    }

    {
      #pragma noinline // inline state is now off
       ...
    }
     // inline state
is back on
    ...
}

In OpenMP, statements that reside within a structured block. See also structured block.

A special type of recurrence that can be parallelized.

See RapidArray link.

The number of links to the file; stored in the file's inode.

The HPC-optimized Linux operating system, based on the SuSE Linux Enterprise Server (SLES) distribution, that runs on each node. Cray optimizations include the implementation of a synchronized process scheduler.

Describing a Cray X1 series system that has a chilled water supply. An LC system uses two chilled-water heat exchange systems: one to cool the Fluorinert liquid, and one to cool the forced air that cools compute modules.

I/O where the records consist of a sequence of values separated by value separators such as commas or spaces. A tab is treated as a space in list-directed input, except when it occurs in a character constant that is delimited by apostrophes or quotation marks.

As it pertains to Psched, a policy wherein applications are distributed among the application nodes. As it pertains to PBS Pro, a policy wherein jobs are distributed across multiple time-shared hosts to even out the workload on each host. The distribution of jobs across execution hosts is solely a function of the PBS Pro Job Scheduler.

A process for ensuring that each processor in a job performs equal work.

The form in which code is presented to the loader. Loader tables, which are generated by compilers and assemblers according to loader requirements, contain information required for loading. This information includes the type of code; names, types, and lengths of storage blocks; data to be stored; and so on.

For UNICOS/mp systems, a collection of culture-dependent information used by an application to interact with a user.

1. Any device or algorithm that is used to ensure that only one process will perform some action or use some resource at a time. 2. A synchronization mechanism that, by convention, forces some data to be accessed by tasks in a serial fashion. Locks have two states: locked and unlocked. 3. A facility that monitors critical regions of code.

An administrator-defined portion of a physical Cray system, operating as an independent computing resource.

An administrator-defined portion of a physical Cray XMT system, operating as an independent computing resource.

A grouping of (Lustre file system) OSTs that are accessed as a unit.

Elapsed time as seen from within processes. This may differ from wall-clock time because it does not include the time processes that are suspended.

The set of all fabric paths in a physical topology. It enables any node in the system to communicate with any other node.

An additional identifying number for a RAID device; the number is used to address regions within the device. LUNs can be further subdivided into partitions.

The service node that provides a user interface and services for compiling and running applications.

The service node that provides a user interface and services for compiling and running applications.

An optimization that combines loop interchange and loop fusion to convert a loop nest into a single loop, with an iteration count that is the product of the iteration counts of the original loops.

An integer variable that is incremented or decremented by an integer constant expression on each pass through a loop.

A form of loop unrolling in which the loop is completely unrolled, and references to loop index within the unrolled loop are replaced by constant values.

An optimization that takes the bodies of loops with identical iteration counts and fuses them into a single loop with the same iteration count.

An optimization that changes the order of loops within a loop nest, to achieve stride minimization or eliminate data dependencies.

A value that does not change between iterations of a loop.

An optimization that increases the step of a loop and duplicates the expressions within a loop to reflect the increase in the step. This can improve instruction scheduling and reduce memory access time.

A code optimization technique in which a loop is unrolled so effectively that it is no longer a loop. For example, a loop with four iterations, each containing a single executable statement, would be replaced by the four executable statements.

Any communication or storage conflict between iterations of a loop. A communication conflict occurs when one iteration writes to a memory location that is read by a another iteration. A storage conflict occurs when two iterations write to the same memory location. The order of the memory operations indicates the nature of the conflict: write-read (flow conflict), read-write (anti conflict), or write-write (output conflict).

Read-write and write-write conflicts are often resolvable by data privatization (that is, assigning distinct storage locations to objects that may be accessed via conflicting iterations). Write-read conflicts are inherent to computation, require explicit synchronization for correct parallel execution of the loop that enclose the conflicts.

The value from one iteration of a loop is used during a subsequent iteration of the loop. This type of loop cannot be parallelized by the compiler.

A listing that is generated by invoking either the Cray Fortran Compiler with the -rm option or the Cray C/C++ Compiler with the -h list=m option. The loopmark listing displays what optimizations were performed by the compiler and tells you which loops were vectorized, streamed, unrolled, interchanged, and so on.

A listing that is generated by invoking the Cray Fortran Compiler with the -rm option. The loopmark listing displays what optimizations were performed by the compiler and tells you which loops were vectorized, streamed, unrolled, interchanged, and so on.

For Cray X1 series systems, a memory chip. Each M-chip controls half of four memory daughter cards (MDCs).

The primary circuit board in a Cray XD1 chassis. The main board has a management processor, RapidArray fabric components, a 10/100 Ethernet switch, three high-speed I/O slots, and connectors for the compute blades, the expansion modules, and the fabric expansion card.

A software service for which the Active Manager software is completely responsible. Includes Active Manager components, DNS, and the LDAP proxy.

The processor on the main board of a Cray XD1 chassis that runs the Active Manager hardware supervisory subsystem.

The chassis that contains the master node.

The node on which the Active Manager server runs.

The thread that creates a team of threads when an OpenMP parallel region is entered.

A decimal multiple equal to 106 bytes (1,000,000 bytes).

For Cray X1 series systems, the hardware unit containing a memory board that is pluggable on a compute module. An MDC contains four channels, each containing four Rambus dynamic random access memory (RDRAM) devices. The interface to the daughter cards is accomplished through the Rambus application-specific integrated circuit (ASIC) cell (RAC).

The hardware unit containing a memory board that is pluggable on a compute module. An MDC contains one Weaver chip drive that drives 20 DDR II SDRAM memory packages, arranged as four channels of 5 packages.

A collection of M-chips and memory daughter cards (MDCs) on a Cray X1 series compute module.

A register that holds status or control information. It is located on M-chips, E-chips, and I-chips.

A widely accepted standard for communication among nodes that run a parallel program on a distributed-memory system. MPI is a library of routines that can be called from Fortran, C, and C++ programs.

A Cray product that consists of the Message Passing Interface and shared distributed memory (SHMEM) data-passing routines.

The component of the Lustre file system that manages Metadata Targets (MDT) and handles requests for access to file system metadata residing on those targets.

A binary multiple equal to 2²⁰ bytes (1,048,576 bytes).

See blade.

A metafile that defines information specific to an application, a collection of applications, or a library. This term is not related to the Fortran language MODULE statement, but is related to setting up the Cray programming environment. The Modules utility uses module files to define the paths, command names, and other environment variables used by the application or library, including the version to be used. For example on Cray XT systems, to set the paths and environment variables required in order to use the default version of the Message Passing (MPI) Toolkit, enter module load xt-mpt to load the xt-mpt module file.

For Cray X1 series systems, an arbitrary number (address) that is assigned to each compute module by initialization software to route packets between compute modules.

A Fortran subroutine or function subprogram. The procedures they define are called module procedures. Module subprograms can be invoked by other module subprograms in the module or by other program units that access the module.

A package on a Cray system that enables you to modify the user environment dynamically by using module files. (This term is not related to the module statement of the Fortran language; it is related to setting up the Cray system environment.) The user interface to this package is the module command, which provides a number of capabilities to the user including loading a module file, unloading a module file, listing which module files are loaded, determining which module files are available for use, and others. For example, the module command can be used to load a specific compiler and its associated libraries, or even a particular version of a specific compiler.

An attribute of the Cray XD1 system (either hardware or software) that the Active Manager software can sense and report. The Active Manager generates alarms based on the conditions of monitored attributes.

An Active Manager agent that collects data from one component of the system.

One of two types of application modes for UNICOS/mp systems. Programs are compiled either as MSP-mode applications (default) or SSP-mode applications. MSP-mode applications run on one or more MSPs. For MSP-mode applications, each MSP coordinates the interactions of its associated four SSPs. See also SSP mode.

Where there are one or more liquid-cooled Cray X1 series cabinets or multiple air-cooled Cray X1 series cabinets, the switch between the CWS and the switches in the Cray X1 series cabinets and the I/O cabinets (IOCs). The top of two switches physically located in the first IOC. Referred to in the Cray X1 series configuration file as topswitch (top switch).

For Cray X1 series systems, the physical packaging that contains processor chips and cache chips. The chips implement either one multistreaming processor (Cray X1 MCM) or two multistreaming processors (Cray X1E MCM). See also MSP.

A processor that combines multiple independent execution engines ("cores"), each with its own cache and cache controller.

Connects Cray Workstations to the system when there are more than two CWSs in the same Cray X1 series system.

A mode that can be set at compile time that ensures that when the compiled application is run, iterations of a loop are run on multiple processors.

For UNICOS/mp systems, a basic programmable computational unit. Each MSP is analogous to a traditional processor and is composed of four single-streaming processors (SSPs) and E-cache that is shared by the SSPs. See also node; SSP; MSP mode; SSP mode.

The concurrent use of multiple threads of control that operate within the same address space.

For UNICOS/mp systems, system operation when there are multiple login sessions, many files open at once, and many active processes, including numerous daemons.

A first-in, first-out file that allows communication between two unrelated processes running on the same host.

I/O that allows you to group variables by specifying a namelist group name. On input, any namelist item within that list may appear in the input record with a value to be assigned. On output, the entire namelist is written.

An IEEE floating-point representation for the result of a numerical operation that cannot return a valid number value; that is, not a number, NaN.

An OpenMP parallel region that appears within a dynamic extent of an OpenMP PARALLEL construct that does not have an if clause or has an if clause that evaluates to true. See also dynamic extent.

Configuration information that an administrator provides to specify a site network to which the Cray XD1 system connects and a range of IP addresses from the network that can be assigned to the system. The Active Manager software uses the network definition in conjunction with partition-network associations and NIC port rules to automatically configure IP interfaces on nodes.

A file system that is exported from a remote server and mounted on other hosts across a network.

A network-based information service and an administrative tool. It allows centralized database administration and a distributed lookup service. NIS supports multiple databases based on regular text files. For example, NIS databases can be generated from the hosts, passwd, group, and aliases files on the NIS master.

For Cray X1 series systems, the protocol that allows synchronizing the clocks in the System Controller and Cray mainframe by pointing the mainframe clocks to the CWS. NTP is used by the CWS to keep the Cray L1 controllers synchronized. This provides consistent time stamps among the various logs for use in diagnosing system problems.

A rule that an administrator defines for the use of a Gigabit Ethernet port. Specifies that a site network always connects to a particular port position. The Active Manager software uses a NIC port rule in conjunction with network definitions and partition-network associations to automatically configure IP interfaces on nodes. A NIC port rule applies to the specified port position on all chassis in the system.

An instance of the Linux operating system and the hardware components that it controls. The hardware components in a Cray XD1 node include an SMP and its associated memory, one or two RapidArray processors (depending on configuration) and, optionally, an FPGA application acceleration processor.

For CLE systems, the logical group of processor(s), memory, and network components acting as a network end point on the system interconnection network.

For UNICOS/mp systems, the logical group of four multistreaming processors (MSPs), cache-coherent shared local memory, high-speed interconnections, and system I/O ports. A Cray X1 system has one node with 4 MSPs per compute module. A Cray X1E system has two nodes of 4 MSPs per node, providing a total of 8 MSPs on its compute module. Software controls how a node is used: as an OS node, application node, or support node. See also compute module; MCM; MSP, node flavor; SSP.

In networking, a processing location. A node can be a computer (host) or some other device, such as a printer. Every node has a unique network address.

For UNICOS/mp systems, software controls how a node is used. A node's software-assigned flavor dictates the kind of processes and threads that can use its resources. The three assignable node flavors are application, OS, and support. See also application node; OS node; support node; system node.

A decimal number used to reference each individual node. The node ID (NID) can be mapped to a physical location.

The hardware-based identifier of a node, which has the following form: "chassis-id.node-ordinal" where "chassis-id" is the chassis ID and "node-ordinal" is the number of the node within the specified chassis (nodes are numbered from 1 to 6, left to right as viewed from the front of the chassis).

See compute module.

The software image that is configured for an individual node and from which the node boots; generated automatically from the partition master software image by the Active Manager software when the node is allocated to a partition. The node working software image is stored either on the local disk of the node or in the Active Manager repository and NFS-mounted.

A command that does not wait for the requested action to complete before it terminates. A few Active Manager commands offer this behavior as an option.

The UNICOS/MP parallel-processing memory architecture. For distributed-memory multinode applications, each application has access to the memory of the other nodes on which it is running. Contrast to uniform memory access (UMA), where threads of shared-memory appplications run on a single node and have uniform access to the node's memory.

The vector instruction set architecture (ISA) for Cray X1 series systems. It supports 32-bit and 64-bit two's-complement integers, IEEE-754 data formats and IEEE-754 style operations, fixed 32-bit width instructions with regular encoding, and decoupled scalar and vector execution. It has large register sets, cache allocation control, and relaxed memory ordering rules with hooks for explicit synchronization.

The component of the Lustre file system that manages Object Storage Targets and handles I/O requests for access to file objects residing on those targets.

The Lustre system component that represents an I/O device containing file data as file system objects. This can be any LUN, RAID array, disk, disk partition, etc.

For a partition: The setting of the access control attribute of a partition that permits end users to log in or submit jobs, depending on the partition's characteristics.
For a node: The setting of the access control attribute of a node that permits end users to log in or submit jobs, depending on the characteristics of its partition.

An industry-standard, portable model for shared memory parallel programming.

For UNICOS/mp systems, the node that provides kernel-level services, such as system calls, to all support nodes and application nodes. See also node; node flavor.

In the Cray XD1 system, use of a communication channel that is separate from the RapidArray interconnect. The supervisory network that monitors and manages the system is an out-of-band Ethernet network.

For Cray X1 series systems, a buffer that contains one of the following entries to indicate what to do with a data request from E-cache: Read--the data returned should be placed into D-cache and the requested word sent to a register; ReadUC--the result should be sent only to a register, not written into D-cache; or Prefetch--the result should be written to D-cache but not returned to a register.

The nonstandard practice of referencing an array with a subscript not contained between the declared lower and upper bounds of the corresponding dimension for that array. This practice sometimes, but not always, leads to referencing a storage location outside of the entire array.

For Cray X1 E systems, a P+ chip consists of 2 processor cores, each containing a scalar section, a vector section, and a memory interface section. P+ chips communicate with system memory through E+ chips. See also P-chip.

For Cray X1 systems, a single processor chip, consisting of a scalar section, a vector section, and a memory interface section. P-chips communicate with system memory through E-chips. See also P+ chip.

The unit of memory addressable through the Translation Lookaside Buffer (TLB). For a UNICOS/mp system, the base page size is 65,536 bytes, but larger page sizes (up to 4,294,967,296 bytes) are also available.

Processing in which multiple processors work on a single application simultaneously.

See serial region.

The software image associated with a partition; used to generate the working software images of nodes that are allocated to the partition. The partition master software image is created from a combination of an application release master, a configuration determined by the partition's attributes, any other partition-wide configuration (such as services), and any installed local or third-party software.

Configuration information that an administrator provides to specify a site network to which the nodes in a partition have access. The Active Manager software uses the partition-network association in conjunction with network definitions and NIC port rules to automatically configure IP interfaces on nodes.

Configuring a UNICOS/mp system into logical systems (partitions). Each partition is independently operated, booted, dumped, and so on without impact on other running partitions. Hardware and software failures in one partition do not affect other partitions.

The route that a RapidArray packet traverses through a specific set of links and switches when it is sent from one endpoint to another endpoint.

On Cray X1 series systems, a logical disk driver that can manage one or more physical paths to each disk logical unit (LUN).

For Cray X1 series systems, the peripheral cabinet that can contain either RAID C-bricks and RAID S-bricks or a Cray Programming Environment Server (CPES), peripherals for the CPES, and the Cray Network Subsystem (CNS).

Optional hardware in the Cray XD1 chassis that connects to one of the slots on the PCI-X expansion card. Supported PCI-X card types include Gigabit Ethernet cards for site network access and Fibre Channel cards for SAN access.

Hardware in the Cray XD1 chassis that provides four PCI-X slots for Gigabit Ethernet and Fibre Channel cards. This card also provides connectors for three disk blades. It connects to one of the three high-speed I/O slots on the main board.

The script that executes the same CrayPat requests that you can enter in line mode.

The text-based user interface to CrayPat.

A set of one or more sections of code that the stream executes in parallel. Each section contains an iteration of a loop. Phases and sections are contained in control flow code generated by the compiler to control the parallel execution of a function.

The physical configuration of the RapidArray interconnect in a Cray XD1 system; implemented by cables that connect RapidArray ports on different chassis either directly or via external RapidArray switches. The system supports two types of physical topology: direct connect and fat tree.

I/O that uses named pipes; faster than normal I/O because it requires less memory than memory-resident files. See also named pipe.

A data item that consists of the address of a desired item.

A low-level message passing interface that enables scalable, high-performance communication between nodes. Applications communicating at the user level link to the Cray MPICH2 or Cray SHMEM library, which are implemented on top of the Portals layer. The Portals interface is transparent to the application programmer.

All the fabric components that the fabric manager has found through periodic or manually invoked fabric sweeps.

The Cray X1 series system internal Ethernet network that connects the Cray Workstation to the Cray Programmer Environment Server and the Cray Network Subsystem. This network provides access to the hardware that is used to configure and operate the programming environment and a site's private network.

A variable that is accessible to only one thread in a team of an OpenMP parallel region.

See access privileges.

A processing element is one instance of an executable propagated by the Application Level Placement Scheduler (ALPS).

The smallest physical compute group. There are two types of processing elements: a compute processing element consists of a Threadstorm processor, memory, and a link to a Cray SeaStar chip. A service processing element consists of an AMD Opteron processor, memory, a link to a Cray SeaStar chip, and PCI-X links.

A hardware element that contains the address of the instruction currently executing.

The UNICOS/mp application placement scheduling tool. The psched command can provide job placement, load balancing, and gang scheduling for all applications placed on application nodes.

A processor that combines four independent execution engines ("cores"), each with its own cache and cache controller, on a single chip.

Eight bytes (64 bits).

For Cray X1 series systems, a router chip, contained within a router module, used to build the Cray X1 series interconnect network. See also router module.

The Cray X1 series internal Ethernet that connects the CWS and the RAID subsystem. The Cray Storage Management (CRAYSM) software uses this network to configure and monitor the RAID subsystem.

The number of dimensions in a Fortran array. Rank is declared when the array is declared and cannot change.

The high-speed network that interconnects the nodes in a Cray XD1 chassis, and connects all nodes in a Cray XD1 system via cables and optional external RapidArray switches. The RapidArray interconnect consists of a main and an optional expansion fabric, each with its own set of fabric components. The configuration of the RapidArray interconnect in a multichassis system is called the physical topology.

The physical communication path between two RapidArray ports. Each link can carry two gigabytes per second.

The special-purpose processor on a Cray XD1 compute blade that is responsible for most communication functions within the system. The RapidArray processor interfaces an Opteron processor to the RapidArray fabric.

A full-crossbar nonblocking switch in the RapidArray fabric. The base configuration includes one 24-port RapidArray switch in each Cray XD1 chassis. The optional fabric expansion card adds a second RapidArray switch. Equivalent external RapidArray switches are available for implementing fat tree (switched) topologies.

An IP core for the FPGA application acceleration processor that provides the logic necessary for an FPGA design to interface (via the RapidArray fabric) to the rest of the Cray XD1 system.

A method of I/O in which the programmer handles the I/O control. See also unformatted I/O.

The time period in which Lustre servers wait for previously connected clients to reconnect. After the recovery window, the service will be available on either the restarted primary server or the backup server.

A recurrence occurs when a loop uses values computed in one iteration in subsequent iterations. These subsequent uses of the value imply loop-carried dependences and thus usually prevent parallelization. To increase parallelization, use linear recurrence.

A simple form of recurrence that reduces a large amount of data to a single value. It is commonly used to find the minimum and maximum elements of a vector. Although similar to a reduction, it is easier to parallelize and uses less memory.

The process of transforming an expression according to certain reduction rules. The most important forms are beta reduction (application of a lambda abstraction to one or more argument expressions) and delta reduction (application of a mathematical function to the required number of arguments). An evaluation strategy (or reduction strategy) determines which part of an expression to reduce first. There are many such strategies. Also called contraction.

A loop that contains at least one statement that reduces an array to a scalar value by doing a cumulative operation on many of the array elements. This involves including the result of the previous iteration in the expression of the current iteration.

A region is an area in code where threads are forked in order to perform a parallel operation. The region ends at the point where the threads join back together at the end of the parallel operation.

A software service that is recorded in the service registry.

An instance of an infrastructure release package or application release package that is installed in the corresponding repository on a Cray XD1 system.

A unit in which Cray delivers software and firmware upgrades for the Cray XD1 system. The two main types of release package are the application release package and the infrastructure release package. Less frequently, the boot release package may change.

System hardware and software design that achieves increased system availability by avoiding or recovering from component and system failures.

A Sun server management tool that allows you to monitor and control the CPES over modem lines and over a network.

See software repository.

A communications interface between the operating environment and the HSS. Each RCA provides an interface between the HSS and the processes running on a node and supports event notification, informational messages, information requests, and probes. See also Hardware Supervisory System (HSS).

To bring a replacement hardware component into service following hardware repair.

For Cray X1 series systems, the physical hardware that connects compute modules.

One of two ways the system converts virtual memory references to physical memory references; the other way is using TLBs (Translation Lookaside Buffers). For multinode applications, the translation of memory references depends on the execution mode. Those run in accelerated mode use an RTT to convert remote virtual memory references to physical memory references. Those run in flexible mode use the local TLB for all memory references; remote address translation is handled via RTT emulation in the TLB. For single-node applications, a TLB converts the application's virtual memory references into physical memory references. TLBs and RTTs are hardware translation functions and therefore faster than the software emulation of RTTs.

See system state.

See system state.

For Cray X1 series systems, the modular, scalable storage component of a RAID subsystem, also called a storage brick. It is located in a PC-20. Each S-brick contains a number of Fibre Channel disk drives. The number of drives per S-brick and the size of each disk varies with the different S-brick modules.

A type of CrayPat experiment that captures values from the call stack or program counter at specified intervals or when a specified event such as a hardware counter overflow occurs.

An operation that uses one or two operands and produces one result.

A form of fine-grain serial processing whereby iterative operations are performed sequentially on the elements of an array, with each iteration producing one result.

The register that serves as source and destination for operands that use scalar arithmetic and logical instructions.

See gather/scatter.

A set of one or more regions collected into a named group for scheduling purposes. Each scheduling domain is controlled by an instantiation of placement, load balancing, and gang scheduling components. Application nodes must be included in a region assigned to one of the scheduling domains if they are to be utilized by Psched.

Part of a program in which a name has a fixed meaning. A program unit or subprogram generally defines a scoping unit. Type definitions and procedure interface bodies also constitute scoping units. Scoping units do not overlap, although one scoping unit may contain another in the sense that it surrounds it. If a scoping unit contains another scoping unit, the outer scoping unit is referred to as the host scoping unit of the inner scoping unit.

An array-processing loop used to perform a table lookup or to find exceptional values within an array.

An area within a program in which only the master task is executing. Its opposite is a parallel region.

To cause only one thread to execute an OpenMP parallel region. A parallel region is said to be serialized if and only if at least one of the following statements is true: (1) the logical expression in an if clause attached to the parallel directive evaluates to false; (2) the parallel region is a nested parallel region and nested parallelism is disabled; and (3) the parallel region is a nested parallel region and the implementation chooses to serialize nested parallel regions.

A unit of work that one software component provides on behalf of another. In the Cray XD1 system, the Active Manager software uses several common system services such as DNS and e-mail transfer.

See blade.

The database that maintains the global system state.

A node that performs support functions for applications and system services. Service nodes run a version of SLES and perform specialized functions. There are six types of predefined service nodes: login, IO, network, boot, database, and syslog.

Performs support functions for applications and system services such as login, network, I/O, boot, and service database (SDB). Service nodes run a version of CLE.

See processing element.

The Active Manager facility that records managed and unmanaged services in the system and protects the nodes that run them from accidental shutdown.

An attribute of an alarm that specifies the degree to which the condition the alarm affects system function. The normal severity values are critical, major, warning. In addition, the severity of a clearing alarm is shown as cleared.

A library of optimized functions and subroutines that take advantage of shared memory to move data between the memories of processors. The routines can either be used by themselves or in conjunction with another programming style such as Message Passing Interface. SHMEM routines can be called from Fortran, C, and C++ programs.

A loop that is vectorized but that has been determined by the compiler to have trips less than or equal to the maximum vector length. In this case, the compiler deletes the loop to the top of the loop. If the shortloop directive is used or the trip count is constant, the top test for number of trips is deleted. A shortloop is more efficient than a conventional loop.

A product that is covered under the terms and conditions of the license agreement that is included with the shipment. The customer need not add this product to the terms and conditions of their standard license agreement signed with Cray.

The result of modifying shared data or performing I/O by concurrent streams without the use of an appropriate synchronization mechanism. Modifying shared data (where multiple streams write to the same location or write/read the same location) without appropriate synchronization can cause unreliable data and race conditions. Performing I/O without appropriate synchronization can cause an I/O deadlock. Shared data, in this context, occurs when any object may be referenced by two or more single-streaming processors. This includes globally visible objects (for example, COMMON, MODULE data), statically allocated objects (SAVE, C static), dummy arguments that refer to SHARED data and objects in the SHARED heap.

A mode that can be set at compile time that ensures that when the compiled application is run, iterations of a loop are run on a single processor.

For UNICOS/mp systems, a basic programmable computational unit. See also node; MSP; MSP mode; SSP mode.

For UNICOS/mp systems, system operation when only a few processes are active on the system and only a single login is allowed.

An ordinary partition that the administrator uses as a template for creating other partitions.

A DNS server that acts as a subsidiary to a master server. Internally, the Cray XD1 system has a master DNS server and at least one slave server. The recommended configuration includes designating an external DNS server as a slave to the Cray XD1 master.

(Relevant to the Grid Engine WLM system.)

A scheduling unit where a single job process can be executed. By default, each core has 1 slot; therefore, by default, single-core nodes have 2 slots and dual-core nodes have 4 slots.

The number of slots per node is configured by a partition attribute: the maximum number of job processes per CPU. The number of slots available in the node is multiplied by this number (the default is 1).

A directory tree that contains the Cray XD1 Linux operating system, application software, and configuration information that is appropriate for the use of the image. Each partition has a master software image and each allocated node has a working software image.

A Cray customer service form and process that tracks software problems from first report to resolution. SPR resolution results either from a written reply, the release of software containing the fix to the problem, or the implementation of the requested design change.

A location in which multiple versions of a release package of a certain type or patches reside. No changes are allowed to the contents of a release package or patch in the repository. The Active Manager software deploys software and firmware upgrades and patches to their target locations by copying from the software repositories.

The process of setting files on the shared-root file system so that unique files can exist for a node or for a class of nodes.

One of two types of application modes for UNICOS/mp systems. Programs are compiled either as MSP-mode applications (default) or SSP-mode applications. SSP-mode applications run on one or more SSPs. Each SSP runs independently of the others, executing its own stream of instructions. In contrast, compiler options enable the programmer to develop command-mode programs that run on an SSP on the support node. See also MSP mode.

A method of allocating memory for variables used by a called routine during program execution. Variables are reset for each invocation of a subprogram. Stack mode is required for multitasked code.

An element of a stack that contains local variables, arguments, contents of the registers used by an individual routine, a frame pointer that points to the previous stack frame, and the value of the program counter at the time the routine was called. A stack frame is allocated when a reentrant subroutine is entered; it is deallocated on exit.

Frequent stack expansion (overflow) and contraction (underflow).

A sequential list of each currently active routine called by a program and the frame pointer that points to its stack frame.

A high-speed interconnect that serves as a bridge between the Cray XT system’s SeaStar interconnect and the YARC interconnect on Cray 2 compute nodes.

A storage architecture where storage is not connected directly to a server, but instead is installed as part of a dedicated, high-speed network.

An attribute of a partition that determines where node working software images are stored: on the node's local disk or on a remote disk.

A Cray Streaming Directives (CSDs) term for a block of code that is to be divided equally among and executed in parallel by the four SSPs that comprise a single MSP.

The relationship between the layout of an array's elements in memory and the order in which those elements are accessed. A stride of 1 means that memory-adjacent array elements are accessed on successive iterations of an array-processing loop.

A single-processor optimization technique in which arrays, and the program loops that reference them, are split into optimally sized blocks termed strips. The original loop is transformed into two nested loops. The inner loop references all data elements within a single strip, and the outer loop selects the strip to be addressed in the inner loop. This technique is often performed by the compiler to maximize the usage of cache memory or as part of vector code generation.

In Fortran OpenMP, a collection of one or more executable statements with a single point of entry at the top and a single point of exit at the bottom. Execution must always proceed with entry at the top of the block and exit at the bottom with only one exception: the block is allowed to have a STOP statement inside a structured block. This statement has the well-defined behavior of terminating the entire program.

The private 100 Mbps Ethernet network internal to the Cray XD1 system that the Active Manager hardware supervisory subsystem uses to monitor and proactively manage the health of the system. Administrators or Cray field engineers can use the supervisory network to commission the Cray XD1 system or diagnose problems.

For UNICOS/mp systems, the node that is used to run serial commands, such as shells, editors, and other user commands (ls, for example). See also node; node flavor.

See fabric.

A table of symbolic names (for example, variables) used in a program to store their memory locations. The symbol table is part of the executable object generated by the compiler. Debuggers use it to help analyze the program.

In a Cray XD1 system, an SMP is formed from two single- or dual-core Opteron processors and their associated memory. One compute blade holds one SMP. Each chassis contains six compute blades and therefore contains six SMPs. See also node.

The Linux process scheduler customized for the Cray XD1 system. It synchronizes time slots across all nodes in the system and allocates more time slots to computing jobs to maximize application performance. Administrators can enable the LSS on a partition-by-partition basis.

I/O operation during which an executing program relinquishes control until the operation is complete. An operation is not complete until all data is moved.

For UNICOS/mp systems, the function that is used to track the use of processor time, number and type of processors, memory, and I/O to secondary storage and over network connections. It is invoked with the sar command.

A set of buffers in kernel memory used for I/O operations by the operating system. The system cache ensures that the actual I/O to the logical device is well formed, and it tries to remember recent data in order to reduce physical I/O requests. In many cases, however, it is desirable to bypass the system cache and to perform I/O directly between the user's memory and the logical device.

For Cray X1 series systems, the Cray L1 controllers, Cray L0 controllers, one or more Cray Workstations (CWSs), and connecting network equipment.

The Cray X1 series system internal network that connects the Cray Workstation, the various Ethernet switches, and Cray L1 controllers in the Cray X1 series mainframe cabinet(s) and I/O cabinets, providing access to the Cray X1 series system for booting, diagnostics, and other control functions.

A remote support tool that monitors the integrity of the Cray system and alerts analysts and system administrators if abnormal operating conditions occur.

The high-speed network that handles all node-to-node data transfers.

The high-speed network that handles all node-to-node data transfers.

The workstation that is the single point of control for system administration. See also Hardware Supervisory System (HSS).

For UNICOS/mp systems, the node that is designated as both an OS node and a support node; this node is often called a system node; however, there is no node flavor of "system." See also node; node flavor.

The proprietary I/O channel that provides I/O access to the Cray X1 series system by connecting the I-chip to an I/O Channel Adapter on an I/O board.

A group of partitions on the BootRAID (boot root, boot node swap, shared root, boot image, SDB, syslog, UFS, etc.) that make a complete, bootable system.

For UNICOS/mp systems, the initialization state of the system. Sometimes referred to as init state, run state, or run level. Common system states are single-user mode (run level 1) and multiuser mode (run level 2). See also multiuser mode; single-user mode.

The main site network that connects to the Cray XD1 system to provide general access to the system. The system subnet contains the default gateway for the system.

The amount of time that the operating system spends providing services to an application.

In the Active Manager GUI, a user-initiated function that is complex and possibly requires multiple steps. The user invokes tasks by clicking links in the sidebar of the GUI or in the workspace of the Tasks view, and the Active Manager software guides the user through the steps in the workspace of the page.

The process of automatically gathering and transmitting data from components of the Cray XD1 system to the Active Manager server.

A phenomenon that occurs when a fixed quantity of a reusable resource is allocated on a least-recently-used basis but the cycle length of the reuse is larger than the quantity of the resource, thereby ensuring that reuse never occurs. When applied to cache, it can slow the execution of a program.

The active entity of execution. A sequence of instructions together with machine context (processor registers) and a stack. On a parallel system, multiple threads can be executing parts of a program at the same time.

An acronym that represents attributes for argument association. It represents the data type, kind type parameter, and rank of the argument.

A table in the processor that contains cross-references between the virtual and real addresses of recently referenced pages of memory. See also RTT (remote translation table).

A memory delay that occurs when a memory location is referenced and the page that contains that memory location does not have an entry in the appropriate translation lookaside buffer.

See multicabinet switch.

A type of CrayPat experiment that counts some event, such as the number of times a specific system call is executed.

A content addressable memory in the processor that contains cross-references between the virtual and real addresses of recently referenced pages of memory.

A command that a user logged into a Cray X1 series system uses to launch Programming Environment components residing on the CPES. Examples of trigger commands are ftn, CC, and pat_build.

A means for categorizing data. Each intrinsic and user-defined data type has four characteristics: a name, a set of values, a set of operators, and a means to represent constant values of the type in a program.

User FPGA processor. A combining form that occurs in file and directory names; for example, in libufp.a. It is a synonym for the FPGA application acceleration processor.

A node that is not assigned to a partition. Unallocated nodes are shut down.

A file that contains undelimited records. Because it does not contain any record control words, it does not have record boundaries.

Transfer of binary data without editing between the current record and the entities specified by the I/O list. Exactly one record is read or written. The unit must be an external unit.

The operating system for Cray X1 series (Cray X1 and Cray X1E) systems.

An extension of the C programming language designed for high performance computing on large-scale parallel processing machines. UPC provides a uniform programming model for both shared and distributed memory hardware. Other parallel programming models include Message Passing Interface, SHMEM, Co-array Fortran, and OpenMP.

A system in which any memory element can be read from or written to in the same, constant time. All processing elements of a single-node application have uniform access to the node's memory. Contrast to nonuniform memory access (NUMA).

A registered software service that runs on a node in the Cray XD1 system and requires administrator intervention when changes occur (for example, if the node fails).

A single-processing-element optimization technique in which the statements within a loop are copied. For example, if a loop has two statements, unrolling might copy those statements four times, resulting in eight statements. The loop control variable would be incremented for each copy, and the stride through the array would also be increased by the number of copies. This technique is often performed directly by the compiler, and the number of copies is usually between two and four.

A software update installation involves applying an update for a software package already running on your system, between major releases for that package. For example, if you install XYZ 7.0.UP05 on a system that is already running an earlier version of XYZ 7.0, it is considered an update installation. See also upgrade installation.

A software upgrade installation involves moving to the next new major release of a software package. Major releases include multiple new features and increased functionality, along with a complete set of updated documentation. For example, installing XYZ 7.0 on a system that is running XYZ 6.0 is considered an upgrade installation. An upgrade installation is different from an software update installation, which occurs between major software releases and provides an update to currently existing software running on your system. See also update installation.

The browser-based Active Manager GUI for Cray XD1 end users. It presents a subset of the GUI capabilities (primarily for job management), in contrast with the Administrator Portal.

The time accumulated by a user process when it is attached to a processor and executing. When running on a single processor, processor time is a fraction of elapsed time. When the program is running in parallel, processor time is a multiple of elapsed time.

The two sides (V-side and W-side) of a liquid-cooled Cray X1 series cabinet. Each side has two bricks; they are named VC0, VC1, WC0 and WC1, where V or W indicates the side of the cabinet and 0 or 1 indicates the brick number. An air-cooled cabinet effectively has only the V-side. See also brick; brick-pair.

A series of values on which instructions operate; this can be an array or any subset of an array such as row, column, or diagonal. Applying arithmetic, logical, or memory operations to vectors is called vector processing. See also vector processing.

The number of elements in a vector.

A register that holds a mask that is used to perform logical operations on the vector registers. The contents of the vector mask may be loaded from a scalar register, read to a scalar register, or formed by logical operations performed on vector registers.

A form of instruction-level parallelism in which the vector registers are used to perform iterative operations in parallel on the elements of an array, with each iteration producing up to 64 simultaneous results. See also vector.

The register that serves as a source and destination for vector operations.

The combined hardware elements that process a program's vector code.

A loop that performs arithmetic on existing elements of an array and stores the results back into the same array. This type of loop requires both a gather operation from and a scatter operation to the same memory location.

An arithmetic or logical expression that consists of a combination of loop invariants, loop counters, vector array references, scalar temporaries, or a function with a vector version that has a vectorizable expression as an argument. This includes most Fortran intrinsic functions.

The process, performed by the compiler, of analyzing code to determine whether it contains vectorizable expressions and then producing object code that uses the vector unit to perform vector processing.

Software that schedules jobs for execution in a system of networked nodes.

The Cray X1 series warning and control system, run by the x1wacs program, that enables the user to control power and monitor hardware status for one or more mainframe cabinets and one or more associated I/O cabinets.

Program that enables job processes to use the Linux synchronized scheduler (in LSS-enabled partitions). This program also sets the CPU affinity for a job process; in other words, it binds the process to a particular CPU (core). End users include the xd1launcher when they submit a job to the Cray XD1 system.

The standard UNICOS/mp file system; a journaling, mature file system with support for very large individual files and file systems.

A generic client that enables you to mount the Lustre file system on any Linux node.