Performance and Scalability

Abstract

This chapter shows typical performance results for instant restart and instant restore, each with a large buffer pool holding the application’s entire working set and with a small buffer pool that incurs buffer faults during transaction processing. The intention here is not a comprehensive evaluation of performance, scalability, or tradeoffs; the purpose here is to demonstrate that instant restart and instant restore, both based on single-page repair and single-transaction rollback, promise substantially shorter perceived downtimes than possible with traditional ARIES techniques. In other words, instant recovery improves system availability by improving the perceived mean time to repair. It is orthogonal to other techniques that might improve the mean time to failure.