Trace replay based I/O performance studies for enterprise workload migration

摘要

Predicting the performance of an application before migrating and deploying on a target system is a daunting but important task. The performance benchmarking of an application in a new environment is a time consuming and expensive process. An approach that performs performance benchmarking of complex application in a new environment without deploying will be an advantage. We have developed and tested a method for predicting the performance of an IO intensive multithreaded enterprise application workload on target storage systems. Our approach is based on well-known trace and replay method. We extract traces of an IO intensive enterprise workload representing temporal and spatial characteristics (e.g. read and write requests) for a smaller workload. These traces are replayed on the target storage system. Using our previously developed tool called PerfExt [1], we can predict the performance of two realistic web applications with less than 15% error on traditional hard disk drive (HDD) and Solid State Drive (SSD) as storage system on the target platform.