|
Abstract : |
Recent research addressed the importance of optimizing L2 cache utilization in the design of main memory indexes and proposed the so-called cache-conscious indexes such as the CSB+-tree. However, none of these indexes took account of concurrency control, which is crucial for running the real-world main memory database applications involving index updates and taking advantage of the off-the-shelf multiprocessor systems for scaling up the performance of such applications. Observing that latching index nodes for concurrency control (CC) incurs the so-called coherence cache misses on shared-memory multiprocessors thus limiting the scalability of the index performance, this paper presents an optimistic, latch-free index traversal (OLFIT) CC scheme based on a pair of consistent node read and update primitives. An experiment with various index CC implementations for the B+tree and CSB+-tree shows that the proposed scheme shows the superior scalability on the multiprocessor system as well as the performance comparable to that of the sequential execution without CC on the uniprocessor system. 1., |
