cient Locking for Shared Memory Database Systems 1

Signi cant performance advantages can be realized by implementing a database system on a shared memory multiprocessor. An e cient implementation of a lock manager is a prerequisite for e cient transaction processing in multiprocessor database systems. To this end, we advocate adopting an autonomous locking strategy. In autonomous locking, transactions acquire and release locks via operations on shared memory, in contrast to conventional locking where communication with a lock manager is involved. We demonstrate the superior performance and scalability of autonomous locking through benchmarks of a prototype lock manager implemented on a large scale shared memory multiprocessor. For instance, when contention is high (i.e., for hot lock operations), the performance of autonomous locking is nearly an order of magnitude better than conventional locking. The paper also addresses issues related to recovery (assuming independent node failures). 1 Partially supported by the National Science Foundation under grant IRI-9109210