ITR: Static and Dynamic Techniques for Latency Hiding in Data-Intensive Applications

点击翻译按钮获取中文摘要

Many of today's performance-critical applications involve manipulating datasets that are either too large or too rarely used to be reliably found inlocal memory caches or local cache servers. Due to the enormous disparitybetween processor cycle times and disk and network access latencies, theseapplications waste a large fraction of their time waiting for data; as welook to the future, this problem is expected to become even worse. Toovercome this problem, this research will combine aggressive storageprefetching with intelligent cache management to fully hide the data accesslatency while using memory resources intelligently. Program transformationtools and runtime support systems will be developed that collaborate tocustomize memory hierarchy and distributed cache resource management fordata-intensive applications. With application-specific guidance of memory,network and disk resources, it is possible to decrease execution times byorders of magnitude. The ultimate goal of the research is for programs tonever waste time waiting for data, and for programmers to never waste theirvaluable programming time thinking about this performance problem.