Methods and Models for Management of Distributed and Persistent Data

this paper we discuss the important issues related to managing persistent data in a distributed system such as a network of workstations. We review research from several areas including distributed file systems, persistent storage systems, and distributed shared memory systems (DSM). A persistent store is a system, such as an object-oriented database, that stores persistent data structures. The main focus of the paper is on this type persistent data because, in a persistent store, the system deals with issues that in a file system are left to programs, such as fine-grain synchronization, coherency, and recoverability. In the most fundamental sense, data management simply means moving data from memory to memory, between disk and memory, and from node to node in the system. Persistent data is cached in local memory so that programs can read or modify it; modifications are sent back to disk to make them persistent. Thus, the key data-management issues pertain to how this cache is managed. We divided our discussion into two parts. First, in Section 3, we discuss issues related to managing data that is accessed only to read it. Here the central questions are (1) how is data organized on disk and in memory and (2) how is the distributed cache hierarchy organized. We compare page-grain and object-grain caching and show that page-grain caches have many benefits. Alternatives for organizing the cache hierarchy are discussed. Most persistent-storage systems are client-server. In contrast, DSM systems have no server; every node is a client. We discuss both of these approaches and conclude by discussing recent work that seeks to extent the cache hierarchy of client-server systems to include the memory of remote clients. Then, in Section 4, we discuss the additional issues raised wh..