Evaluation of integration of ACBL and AOCC caching algorithms

Transactional caching algorithms proposed for object database management systems (ODBMSs) have been classified into detection and avoidance categories, depending on whether they allow access to stale data. Although studies have shown that in most situations the leading detection-based algorithms tend to outperform those that use avoidance, most ODBMSs use a variation of the leading avoidance-based algorithm because detection algorithms in certain situations lead to abort rates that are unacceptable for some, typically interactive, applications. The Interoperable Server-side Caching (ISCT) algorithm allows both types of algorithms to interoperate in the same environment while being supported by an interoperable server. Clients use either the leading detection-based algorithm, called Adaptive Optimistic Concurrency Control (AOCC), or the leading avoidance-based algorithm, called Adaptive Call-Back Locking (ACBL). The caching operations of the client-side algorithms are not affected - only the server-side is adapted to interoperate with both types of clients. This paper presents the results of performance evaluation that compares the three algorithms under various loads. Evaluation shows that the interoperable server-side caching algorithm increases throughput in environments where some applications can use the high-throughput AOCC algorithm while other applications use a detection-based ACBL algorithm. To compare the algorithms a new workloads, in which different applications exhibit different server load characteristics, were developed.