High Performance Computing and Communications Glossary 2.1

A significant part of the material of this glossary was adapted from material originally written by Gregory V. Wilson which appeared as "A Glossary of Parallel Computing Terminology" (IEEE Parallel & Distributed Technology, February 1993), and is being re-printed in the same author's "Practical Parallel Programming" (MIT Press, 1995). Several people have contributed additions to this glossary, especially Jack Dongarra, Geoffrey Fox and many of my colleagues at Edinburgh and Syracuse.

Original version is from NPAC at <URL:http://nhse.npac.syr.edu/hpccgloss/>

Original author: Ken Hawick, khawick@cs.adelaide.edu.au

See also the index of all letters and the full list of entries (very large)

Sections: A B C D E F G H I J K L M N O P Q R S T U V W X Y Z

R

r-inf (n.) (pronounced r-inf, often written with a subscript) The performance a pipelined architecture would deliver on an infinitely long vector; that is, the performance of such an architecture when startup costs are not considered. This parameter was coined by Hockney and Jesshope in 1988. See also n-1/2.

race condition(n.) A situation in which the final result of operations being executed by two or more processes depends on the order in which those processes execute. For example, if two processes A and B are to write different values VA and VB to the same variable, then the final value of the variable is determined by the order in which A and B are scheduled.

RAID (n.) Redundant array of inexpensive disks; a file system containing many disks, some of which are used to hold redundant copies of data or error correction codes to increase reliability. RAIDS are often used as parallel access file systems, where the sheer size of storage capacity required precludes using more conventional (but more expensive) disk technology.

RAM (n.) Random Access Memory; computer memory which can be written to and read from in any order. See also DRAM, SRAM.

random uniform game tree (n.) a game tree whose terminal node values are randomly chosen from some uniform distribution.

randomized routing (n.) A routing technique in which each message is sent to a randomly chosen node, which then then forwards it to its final destination. Theory and practice show that this can greatly reduce the amount of contention for access to links in a multicomputer.

rank (n.) row of a butterfly network.

RDMS (n.) Relational Database Management System; software to manage a database in which data are stored by attribute.

ready list (n.) OS list containing ready-to-run processes.

reasonable (adj.) a parallel model is said to be reasonable if the number of processors each processor can communicate with directly is bounded by a constant.

recurrence (n.) a relationship in a DO-loop whereby a computation in one iteration of the loop depends upon completion of a previous iteration of the loop. Such dependencies inhibit vectorization.

reduced instruction set computer (adj.) See RISC

reduction operation (n.) An operation applying an associative and commutative binary operator to a list of values, See also parallel prefix.

redundant array of inexpensive disks (n.) See RAID

redundant computation (n.) Calculations that are carried out more than once or by more than one processor. Computations may be done redundantly because it is cheaper to have every processor calculate a value for itself than to have one processor calculate the value and then broadcast it, or because processes may not have enough memory to store all the values they calculate and may need to overwrite some during execution.

refute (v.) to make a move that causes an alpha-beta cutoff.

relaxation method (n.) A type of indirect method in which the values making up a trial solution to an equation are repeatedly updated according to some rule until convergence criteria are met. See also direct method

remote procedure call (n.) a structured implementation of a client-server interaction.

rendezvous (n.) when the server side of a remote procedure call is specified by using an accept statement or similar construct.

reply message (n.) passing of results back to the client in a remote procedure call.

RGB (adj.) Red-Green-Blue; the most common form of colour display hardware.

ring (n.) A topology in which each node is connected to two others to form a closed loop. See also chain, Hamiltonian.

RIP (n.) routing information packet is an IGP supplied with BSD networking Unix.

RISC (adj.) Reduced instruction set computer; a computer that provides only a few simple instructions but executes them extremely quickly. RISC machines typically rely on instruction prefetching and caching to achieve higher performance than CISC machines. The term is also applied to software designs that give users a small number of simple but efficient operations.

ROM (n.) Read Only Memory; a computer memory which cannot be written to during normal operation.

routing (n.) The act of moving a message from its source to its destination. A routing technique is a way of handling the message as it passes through individual nodes. See also e-cube routing, interval routing, Metropolis routing, packet switching, randomized routing, virtual cut-through, wormhole routing.

routing algorithm (n.) a rule for deciding, at any intermediate node, where to send a message next. See also routing.

routing kernel (n.) See kernel.

RPC (n.) remote procedure call is a popular model for implementing distributed client-server computing environments. It is an alternative to inter-process communication (IPC) which allows remote systems to execute a set of procedures to share information.

rule (n.) in the context of logic programming, a rule is a Horn clause with a head and a body.