BibTeX
@MISC{_multi-wordatomic,
author = {},
title = {Multi-word Atomic Read/Write Registers on Multiprocessor Systems},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
1 Introduction In multiprocessing systems cooperating processes may share data via shared data objects. In this paper we are interested in designing and evaluating the performance of ef^cient shared data objects for cooperative tasks in multiprocessor systems. More speci^cally we are interested in designing a practical wait-free algorithm for implementing registers (or memory words) of arbitrary length that could be read and written atomically. (Typical modern multiprocessor systems support words of 64-bit size.) The most commonly required consistency guarantee for shared data objects is atomicity, aka linearizability. A shared object (its implementation) is atomic or linearizable if it guarantees that even when operations overlap in time, each of them appears to take effect in an atomic time instant which lies in its respective time duration, in a way that the effect of each operation is in agreement with the object's sequential speci^cation. The latter means that if we speak of e.g. read/write objects, the value returned by each read equals the value written by the most recent write according to the sequence of S,shrunkT ^ operations in the time axis.
Keyphrases
multiprocessor system multi-word atomic read write register data object arbitrary length share data 64-bit size aka linearizability time axis ef cient atomic time instant sequential speci cation practical wait-free algorithm recent write respective time duration shrunkt operation memory word cooperative task write object required consistency guarantee
