Adaptable Distributed Shared Memory: a Formal Definition

Jordi Bataller and JoseÅ M. Bernabeu-Auban

Abstract
Distributed shared memory (DSM) is a paradigm for programming paralleland distributed systems. It offers the agents of the system a sharedspace address through which they can communicate each other.The main problem of a DSM implementation is performance, speciallywhen it is implemented on top of a message passing systems. This, ofcourse, depends on the consistency the DSM system offers, since strongconsistency is more expensive to implement than weak consistency.Many systems have been proposed, each one supporting a different levelof consistency. Unfortunately, experience has shown that none iswell suited for the whole range of problems. In some cases, strongprimitives are not needed while in others too weak primitives areuseless. This is also true for different implementations of the samememory model, since no single one is able to efficiently support allof the possible data access patterns.In this paper, we introduce a novel model called Mume. Mume is a lowlevel layer close to the level of the message passing interface. Mumeinterface only offers the strictly necessary requirements to beconsidered as shared memory thus allowing an efficient implementation.The interface includes three types of synchronization primitives,namely, total ordering, causal ordering and mutual exclusion.Mume interface can be used either to develop final applications or tobuild new primitives and use them in turn to develop final programs.This latter possibility allows, for example, to emulate the main memorymodels proposed to date. Additionally, the same primitive can implementedwith different behaviours thus permitting to fit different data accesspatterns.This paper formally describes Mume and its properties. We use aframework to reason about memory models previously developed by theauthors. To show Mume adaptability, we prove how Mume can emulate mostof the main memory models.
Contact
Jordi Bataller
Escola Universitaria de Gandia,Ctra. Nazaret-Oliva s/n,46.730 Grau de Gandia, Valencia, SPAIN
bataller@iti.upv.es