The exponential growth in computing capability and use has produced ahigh demand for large, high-performance storage systems.Unfortunately, advances in storage system research have been limitedby (1) a lack of evaluation workloads, and (2) a limited understandingof the interactions between workloads and storage systems.We havedeveloped a tool, the Distiller that helps address bothlimitations.Our thesis is as follows: Given a storage system and a workload forthat system, one can automatically identify a set of workloadcharacteristics that describes a set of synthetic workloads with thesame performance as the workload they model.These representativesynthetic workloads increase the number of available workloads withwhich storage systems can be evaluated.More importantly, thecharacteristics also identify those workload properties that affectdisk array performance, thereby highlighting the interactions betweenworkloads and storage systems.This dissertation presents the design and evaluation of the Distiller.Specifically, our contributions are as follows. (1) We demonstratethat the Distiller finds synthetic workloads with at most 10% errorfor six out of the eight workloads we tested. (2) We also find thatall of the potential error metrics we use to compare workloadperformance have limitations.Additionally, although the internalthreshold that determines which attributes the Distiller chooses has asmall effect on the accuracy of the final synthetic workloads, it hasa large effect on the Distiller's running time. Similarly, (3) we findthat we can reduce the precision with which we measure attributes andonly moderately reduce the resulting synthetic workload'saccuracy. Finally, (4) we show how to use the information contained inthe chosen attributes to predict the performance effects of modifyingthe storage system's prefetch length and stripe unit size.
【 预 览 】
附件列表
Files
Size
Format
View
Generating and Analyzing Synthetic Workloads using Iterative Distillation
PDF
download
878987797
028-85220240
