Abstract Competitive Prefetching for Data-Intensive Online Servers £

Recent studies on operating system support for highly concurrent online servers mostly target CPU-intensive workloads with light disk I/O activities. However, an important class of online applications that access a large amount of disk-resident data, such as the index searching of large-scale Web search engines, has received limited attention. This paper examines operating system techniques to improve the throughput of data-intensive online servers under concurrent execution. Our contributions are two-fold. First, we propose a competitive prefetching strategy that balances the overhead of disk I/O switching and the wasted I/O bandwidth of prefetching unnecessary data. Second, we enhance the operating system memory management to reduce prefetching-incurred page thrashing at high concurrency levels. Our enhancements include the introduction of a new cache dedicated to prefetched pages and a novel page reclamation algorithm for it. We have implemented the proposed techniques in the Linux 2.6.3 kernel and conducted experiments based on microbenchmarks and three real applications (an index searching server, the Apache Web server, and a genetic sequence database). Our evaluation results demonstrate that competitive prefetching can improve the throughput of real applications by up to 47 % at low concurrency while the enhanced memory management scheme can produce up to a four-fold performance enhancement at high concurrency. The performance improvement is achieved without any application changes.