Characterization of the data access behavior for TPC-C traces

Abstract

In this paper, we look into the characteristics of the reference stream of TPC-C workloads from the buffer pool point of view. We analyze a trace coming from DB2 UDB version 8.1 fix pack 4 and compare it to a trace from DB2 UDB version 8.1 GA. We perform three types of analysis. A static analysis of the number of reads and writes for index and data pages. We conclude that index pages receive less references than data pages by are more frequently accessed individually. Then, we analyze how DB2 processes access those pages. Index pages have more references than data pages when accessed by more than one process. Finally, we understand the accesses along the life of a page. We conclude that there is a significant burstiness in the reference stream, where, each burst is caused by one process.