The Exclusive-Writer Approach to Updating Replicated Files in Distributed Processing Systems

Consistency control protocols can be classified as either pessimistic or optimistic. Pessimistic protocols check for conflicting file accesses before a transaction references shared files; this prevents transaction restarts but adds intercomputer synchronization delays to execution response times TE. Optimistic protocols avoid intercomputer synchronization delays for TE, but existing optimistic protocols repeatedly restart a transaction until it executes without conflict. Repeated restarts lengthen the time to finalize an update TU, and can saturate the computing and communication resources. We present two new optimistic protocols that avoid repeated restarts: the exclusive-writer protocol (EWP) and the exclusive-writer protocol with locking option (EWL). EWP has no transaction restarts, database rollbacks, or deadlocks due to shared data access. But EWP ensures only a limited form of serializability. EWL is an extension of EWP that ensures full serializability. EWL has no database rollbacks. Also, EWL can guarantee that a transaction will be restarted at most once. To further reduce restarts, each site can independently and dynamically switch between primary site locking (PSL), which has no restarts, and EWL. Such switching requires no additional messages or delays to synchronize protocol selection. Analytic models are developed to study the response times (i.e., TE and TU) of EWP, EWL, PSL, and basic timestamps (BTS). Our study reveals that EWP and EWL have the smallest TF since neither requires update-log maintenance (unlike BTS) nor intercomputer synchronization delays for TE (unlike PSL). EWP has the smallest TU unless the cost of communicating and processing updates is high.