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


WAN (n.) Wide area network, a network of circuits spanning a large region or global in proportions, that is used to transmit data between widespread subscribers. See also LAN.

WARPS (n.) Words accessed randomly per second; a measure of memory access performance, equal to the rate of uniformly random accesses across the whole of the address space visible to a process that a machine supports. See also FLOPS.

weak search (n.) a search algorithm that searches for a key and returns the node that contained the key at the time it was examined. Weak searches are not guaranteed to provide an up-to-date result. See also strong search.

weight (n.) a real number assigned to an edge in a weighted graph.

weight matrix (n.) a matrix indicating, for each pair of vertices I and J, the weight of the edge from vertex I to vertex J.

weighted graph (n.) a graph with a real number assigned to each edge.

working set (n.) Those values from shared memory that a process has copied into its private memory, or those pages of virtual memory being used by a process. Changes a process makes to the values in its working set are not automatically seen by other processes.

wormhole routing (n.) A technique for routing messages in which the head of the message establishes a path, which is reserved for the message until the tail has passed through it. Unlike virtual cut-through, the tail proceeds at a rate dictated by the progress of the head, which reduces the demand for intermediate buffering. See also packet switching.

worst-case space complexity (n.) greatest amount of space used by an algorithm, over all possible inputs of a given size.

wrap-around scalar (n.) a scalar variable whose value is set in one iteration of a DO-loop and referenced in a subsequent iteration and is consequently recursive. All common reduction-function scalars are wrap-around scalars and usually do not prevent vectorization. All other wrap-around scalars usually do prevent vectorization of the loop in which they appear.

write-back cache (n.) See write-in cache.

write-in cache (n.) a cache in which writes to memory are stored in the cache and written to memory only when a rewritten item is removed from cache. This is also referred to as write-back cache.

write-through cache (n.) a cache in which writes to memory are performed concurrently both in cache and in main memory.