Study of the Performance Impact of NFS Workload on Large Mainframe UNIX(R) Systems.

This paper studies the performance impact of the workload generated by Network File System (NFS) applications, on Large Mainframe UNIX(R) systems when used as NFS servers. To quantify the load generated by NFS operations, a series of controlled experiments with synthetic and application benchmarks were performed. The benchmarks were designed to study server and client CPU requirements both for synthetic and application NFS workloads. The server resource requirements for NFS write requests were studied in detail and a model was developed to describe the server CPU consumption per write request. The performance impact of NFS application workload, generated by diskless and dataless NFS clients was studied and compared. Based on the results of the experiments several recommendations for system engineering were made.