Avoiding version redundancy for high performance reads in temporal databases

Abstract

A major performance bottleneck for database systems is the memory hierarchy. The performance of the memory hierarchy is directly related to how the content of disk pages maps to the L2 cache lines, i.e. to the organization of data within a disk page, called the page layout. The prevalent page layout in database systems is the N-ary Storage Model (NSM). As demonstrated in this paper, using NSM for temporal data deteriorates memory hierarchy performance for query-intensive workloads. This paper proposes two cacheconscious, read-optimized, page layouts for temporal data. Experiments show that the proposed page layouts are substantially faster than NSM.