Parallel query processing in shared disk database systems

Abstract

System developments and research on parallel query processing have concentrated either on “Shared Everything” or “Shared Nothing” architectures so far. While there are several commercial DBMS based on the “Shared Disk” alternative, this architecture has received very little attention with respect to parallel query processing. A comparison between Shared Disk and Shared Nothing reveals many potential benefits for Shared Disk with respect to parallel query processing. In particular, Shared Disk supports more flexible control over the communication overhead for intra-transaction parallelism, and a higher potential for dynamic load balancing and efficient processing of mixed OLTP/query workloads. We also sketch necessary extensions for transaction management (concurrency/coherency control, logging/recovery) to support intra-transaction parallelism in the Shared Disk environment.