Report ID
1994-09
Report Date
Abstract
Although the two-phase commit (2PC) protocol has been studied extensively for along time, the topic is of sufficient significance that several refinements ofthe protocol have continued to emerge. However, much of the existingliterature on 2PC is restricted to discussing and analyzing the protocol (andits variants) in the absence of failure. Very little, especially inquantitative terms, has been written about its performance in the presence ofsite failures. In this study, we use a simulation testbed of a distributeddatabase system to quantify the differences in the performances of four 2PCprotocols. Our study covers both the no-failure case and the case of sitefailures. We present a number of interesting experiment results. One is thatthe performance of these protocols is highly dependent on themessage-processing latency at the transaction coordinator site. Another isthat the presumed abort protocol does not necessarily yield better performancein the presence of site failures.
Document
1994-09.ps170.88 KB