DiskAccel: Accelerating Disk-Based Experiments by Representative Sampling

摘要: Disk traces are typically used to analyze real-life workloads and for replay-based evaluations. This approach benefits from capturing important details such as varying behavior patterns, bursty activity, diurnal patterns of system which often missing the workload synthesis tools. However, accurate capture requires recording containing long durations difficult use evaluation. One way solving problem storage trace duration is disk simulators. While publicly available simulators can greatly accelerate experiments, they have not kept up with technological innovations in field. The variety, complexity, opaque nature hardware make it very implement alternative, replaying whole on real hardware, suffers either run-time or required manual reduction experimental time, potentially at cost reduced accuracy. On other hand, burstiness, auto-correlation, complex spatio-temporal properties known methods sampling less effective. In this paper, we present a methodology called DiskAccel efficiently select key intervals representatives replay them estimate response time workload. Our extracts variety spatial temporal features each interval uses efficient data mining techniques representative intervals. To verify proposed methodology, implemented tool capable running selective warming state an accelerated manner, emulating request causality while minimizing inter-arrival error. Based our manages speed by more than two orders magnitude, keeping average estimation error 7.6%.