Mutually Consistent Recording in Asynchronous Computations
نویسندگان
چکیده
A global record (i.e. a set of local records, one for each process of an asyn-chronous computation) abstracts what is usually called global state, global checkpoint or global snapshot in particular problems. This paper concentrates on consistency of global records. First, a general model of asynchronous computations, including the classical shared memory model and several message passing models, is introduced. Then, under this general model, a necessary and suucient condition, stating whether an arbitrary set of local records can be included in some consistent global record, is proved. Finally, it is shown that, when a simple strategy (derived as a consequence of the previous theorem) is followed by each process of an asynchronous computation, then all local records taken by processes belong to consistent global records. Such a result can be used to design eecient snapshoting or checkpointing algorithms in asynchronous computations. the HCM project \CABERNET" (3702) and the ESPRIT BRA project "BROADCAST" (6360). Accepted as short presentation in 15th ACM Symposium on Principles of Distributed Computing (PODC'96, may 1996, Philadelphia) under the title "About recording of asynchronous computation". Enregistrements mutuellement coh erents dans les calculs asynchrones R esum e : Un enregistrement global (c'est-a-dire un ensemble d'enregistrements locaux, un pour chaque processus d'un calcul asynchrone) recouvre les notions ha-bituelles d' etat global, point de contr^ ole global, ou encore instantan e, rencontr ees dans dii erents probl emes particuliers. On introduit d'abord un mod ele g en eral de calcul asynchrone, englobant les mod eles classiques de m emoire partag ee et de communication par messages. Dans ce cadre g en eral, on etablit ensuite une caract erisation des ensembles d'enregistre-ments locaux pouvant faire partie d' enregistrements globaux coh erents. Ennn, on montre que si tous les processus d'un calcul asynchrone appliquent une certaine strat egie (d eduite simplement de la caract erisation pr ec edente), alors tous les enre-gistrements locaux peuvent appartenir a des enregistrements globaux coh erents. Une telle strat egie peut ^ etre utilis ee pour construire des algorithmes eecaces de calcul d'enregistrements globaux coh erents dans les calculs asynchrones.
منابع مشابه
DALD:-Distributed-Asynchronous-Local-Decontamination Algorithm in Arbitrary Graphs
Network environments always can be invaded by intruder agents. In networks where nodes are performing some computations, intruder agents might contaminate some nodes. Therefore, problem of decontaminating a network infected by intruder agents is one of the major problems in these networks. In this paper, we present a distributed asynchronous local algorithm for decontaminating a network. In mos...
متن کاملImage Analysis with R-operators
Associative nets is a well suited parallel computing model for image analysis. It allows to perform asynchronous computations an irregular data. Some e cient implementations have been designed, either in hardware or software. Asynchronism can largely improve the algorithm speeds, provided these algorithms can be written with operators that allow the asynchronous implementation. Classical operat...
متن کاملPerspectives on asynchronous computations for fluid flow problems
Instances where asynchronous parallel computations can be used for the solution of fluid flow problems are discussed. The use of asynchronous portions of solvers can effectively counteract the consequences of the inefficiencies associated with high latency in the interprocessor (or intercomputer) communication. This is especially so in the case of irregular regions, irregular partitions, or in ...
متن کاملSynchronizers for Local Computations
A synchronizer is intended to allow synchronous algorithms to be executed on asynchronous networks. It is useful because designing synchronous algorithms is generally much easier than designing asynchronous ones. In this paper, we provide synchronization protocols described as local computations. We obtain a general and an unified approach for handling synchrony in the framework of local comput...
متن کاملDistributed Asynchronous Custom DSP Systems Applied on Successive Interference Cancellation
To combat clock distribution problems, locally synchronous, globally asynchronous methods are of great interest. As an application example of this, a prestudy for a custom DSP implementation of a multiuser detector scheme based on non-decision directed interference cancellation is presented. Two architectural implementation methods for asynchronous detection are suggested and mutually compared.
متن کامل