PROCESSING TRANSACTIONS WITH GREATER ACCURACY AND AVAILABILITY IN COPAR

摘要

This paper investigates the accuracy and availability of COPAR, a distributed service that integrates optimistic and pessimistic methods for managing updates to counter like replicas at a predetermined, tolerable, penalty. The service utilizes a variable, system cost bound, which captures the level of inconsistency the system can bear. Optimistic processing increments and decrements a server's share of the cost bound as dictated by transactions while pessimistic processing calculates a new system cost bound every time a transaction is committed and the new quantity is subsequently distributed among the servers. It is demonstrated that the use of the formula previously applied for calculating the new system cost bound whenever a transaction is committed may result in unacceptable, over allocation of resources and a simple alternate formula that reduces such potentially expensive inaccuracies is presented. The service runs over the Internet and there are different classes of failure that may affect the servers and communication links involved and modest attention has been given to these failure types in the context of the combination concurrency control technique. This paper outlines modifications to server behaviour in COPAR that enhance availability.