2000•Unpublished venueRequires access

On the Verification of Memory Models of Shared-Memory Multiprocessors

Shaz Qadeer

Open publisher page 10 citations

Abstract

The memory model of a shared-memory multiprocessor is a contract between the designer and programmer of the multiprocessor. We present a model checking algorithm to verify this contract for finite values of the parameters ---number of processors and number of memory locations--- for a large class of shared-memory systems and memory models. A memory model is a generalization of serial memory, which behaves as if there is a centralized memory that services read and write requests atomically such that a read to a location returns the latest value written to that location. We formalize a memory model as an irreflexive partial order among the memory (read and write) events performed locally at each processor. A run of a memory system satisfies a memory model if there exists a total order of all memory events that is both consistent with all the partial orders and a trace of serial memory. Sequential consistency is an example of a well-known memory model. It has been shown that even for fini...

About this research paper

What this paper is about

The memory model of a shared-memory multiprocessor is a contract between the designer and programmer of the multiprocessor. We present a model checking algorithm to verify this contract for finite values of the parameters ---number of processors and number of memory locations--- for a large class of shared-memory systems and memory models. A memory model is a generalization of serial memory, which behaves as if there is a centralized memory that services read and write requests atomically such that a read to a location returns the latest value written to that location. We formalize a memory model as an irreflexive partial order among the memory (read and write) events performed locally at each processor. A run of a memory system satisfies a memory model if there exists a total order of all memory events that is both consistent with all the partial orders and a trace of serial memory. Sequential consistency is an example of a well-known memory model. It has been shown that even for fini...

Why it matters

OpenAlex reports 10 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

The memory model of a shared-memory multiprocessor is a contract between the designer and programmer of the multiprocessor. We present a model checking algorithm to verify this contract for finite values of the parameters ---number of processors and number of memory locations--- for a large class of shared-memory systems and memory models. A memory model is a generalization of serial memory, which behaves as if there is a centralized memory that services read and write requests atomically such that a read to a location returns the latest value written to that location. We formalize a memory model as an irreflexive partial order among the memory (read and write) events performed locally at each processor. A run of a memory system satisfies a memory model if there exists a total order of all memory events that is both consistent with all the partial orders and a trace of serial memory. Sequential consistency is an example of a well-known memory model. It has been shown that even for fini...

Key concepts: Memory model, Computer science, Flat memory model, Memory map, Shared memory, Distributed shared memory, Distributed memory, Extended memory

Related papers

Back to paper searchBrowse research topicsOriginal source
On the Verification of Memory Models of Shared-Memory Multiprocessors — Research Paper | ScholarLens