Reducing the blocking in two-phase commit protocol employing backup sites

In distributed database systems (DDBSs), a transaction blocks during two-phase commit (2PC) processing if the coordinator site fails and at the same time some participant site has declared itself ready to commit the transaction. The blocking phenomena reduces availability of the system since the blocked transactions keep all the resources until they receive the final command from the coordinator after its recovery. To remove the blocking problem in 2PC protocol, the three phase commit (3PC) protocol was proposed. Although 3PC protocol eliminates the blocking problem, it involves an extra round of message transmission, which further degrades the performance of DDBSs. We propose a backup commit (BC) protocol by including a backup phase to 2PC protocol. In this, one backup site is attached to each coordinator site. After receiving responses from all participants in the first phase, the coordinator communicates its decision only to its backup site in the backup phase. Afterwards, it sends a final decision to participants. When blocking occurs due to the failure of the coordinator site, the participant sites consult the coordinator's backup site and follow termination protocols. In this way, BC protocol achieves a non-blocking property in most of the coordinator site failures. However, in the worst case, the blocking can occur in BC protocol when both the coordinator and its backup site fail simultaneously. If such a rare case occurs, the participants wait until the recovery of either the coordinator site or the backup site. BC protocol suits DDBS environments in which sites fail frequently and messages take longer delivery time. Through simulation experiments it is shown that BC protocol exhibits superior throughput and response time performance over 3PC protocol and performs closely with 2PC protocol.