Adding Response Time Measurement of
CIFS File Server Performance to NetBench

Karl L. Swartz - Network Appliance

Abstract

The standard benchmark for NFS file server performance, SPEC SFS (also known as LADDIS), measures performance in terms of both throughput -- the aggregate amount of data a file server can move across the network per unit of time -- and response time -- the time required to service an individual client request. NetBench, the most commonly used file server benchmark for the CIFS (or SMB) protocol measures only throughput. Network Appliance believes response time is as important a performance metric as throughput, especially in the highly interactive environment typical of CIFS networks, since throughput offers little solace to a user waiting to access a file.

This paper documents the methodology and tools developed to measure response time during a NetBench run. While cumbersome and primitive, useful data has been produced, demonstrating that the fundamental idea is sound. SPEC SFS has had a noticeable effect on vendors of NFS file servers, motivating them to improve response time from an average of 50ms in 1993 to less than 10ms in 1997. Given the ability to measure response time in the CIFS environment, hopefully a similar improvement can be encouraged in CIFS file servers.

Presented (PPT, HTML) at the USENIX Windows NT Workshop, Seattle, 11-13 August 1997. (Proceedings, HTML, PDF, PS.GZ)

Also presented (PPT, HTML) at the Large Scale System Administration of NT Workshop, Seattle, 14-16 August 1997.

  Karl's Published Papers
  Karl's Home Page
KLSē   KLSē Home
  Network Appliance

Powered by FreeBSD Copyright © 1997-1998,2001-2003, Karl L. Swartz. All rights reserved.
All trademarks mentioned herein belong to their respective owners.

Karl Swartz <karl@kls2.com>