Distributed Concurrent Persistent Languages: An Experimental Design and Implementation

Computer science

Wai, H. W. Francis

University:University of Glasgow

关键词: Computer science;   

Others  :  http://theses.gla.ac.uk/77719/1/10998190.pdf

来源: University of Glasgow

【 摘 要 】

A universal persistent object store is a logical space of persistent objects whose localities span over machines reachable over networks. It provides a conceptual framework in which, on one hand, the distribution of data is transparent to application programmers and, on the other, store semantics of conventional languages is preserved. This means the manipulation of persistent objects on remote machines is both syntactically and semantically the same as in the case of local data. Consequently, many aspects of distributed programming in which computation tasks cooperate over different processors and different stores can be addressed within the confines of persistent programming. The work reported in this thesis is a logical generalization of the notion of persistence in the context of distribution. The concept of a universal persistent store is founded upon a universal addressing mechanism which augments existing addressing mechanisms. The universal addressing mechanism is realized based upon remote pointers which although containing more locality information than ordinary pointers, do not require architectural changes. Moreover, these remote pointers are transparent to the programmers. A language, Distributed PS-algol, is designed to experiment with this idea. The novel features of the language include: lightweight processes with a flavour of distribution, mutexes as the store-based synchronization primitive, and a remote procedure call mechanism as the message-based interprocess communication mechanism. Furthermore, the advantages of shared store programming and network architecture are obtained with the introduction of the programming concept of locality in an unobtrusive manner. A characteristic of the underlying addressing mechanism is that data are never copied to satisfy remote demands except where efficiency can be attained without compromising the semantics of data. A remote store operation model is described to effect remote updates. It is argued that such a choice is the most natural given that remote store operations resemble remote procedure calls.