Workload Characterization of AFS File Servers

Abstract

This paper describes the workload characterization of AFS file servers, based on traces collected by the file servers at CITI over a 2-4 week period. These workload characteristics have been used to compare the performance of servers running on different hardware. There are two parts to this paper. In the first part, we describe a server model that will be used to drive a synthetic workload. In the second part, we build a client model consisting of the requests made by different user types. We show that the user community can be broken into distinct types, where all the users of a specific type exhibit similar request patterns. This clustering information is used in conjunction with the workload characteristics at the server to predict loads on servers, given the community that needs to be served. Finally, we give an example to show how this can be done.