Modeling the performance and energy of storage arrays

摘要: We propose a novel framework for evaluating techniques power optimization in storage. Given an arbitrary trace of disk requests, we split it into short time intervals, extract set simple statistics each interval, and apply analytical model to those obtain accurate information regarding the performance energy characteristics system that workload. The key abstraction used our is run-length - single sequential run requests at level. Using this abstraction, able account interactions random I/Os context RAID array, results with less effort than detailed individual request-level simulation. Various layout migration policies aimed conservation can be easily expressed as transformations on interval. demonstrate efficacy by using evaluate PARAID, recently proposed technique storage arrays. show predicted under PARAID accurately match simulation system. analytic allows us identify parameters affect performance, enhancement data which perform superior original technique. use both simulations illustrate benefit new layout. This also demonstrates significant simplicity applying high-level extracted statistics, compared current alternative either implementing or simulating level