cosc038@canterbury.ac.nz (08/10/90)
An honours student in our department is doing a project to construct
a simulation model of the performance of an NFS file service. Proposed
input parameters for the model include:
- the server configuration - especially the number of disc, and the
performance characteristics of the discs, the size of the
server's memory, and the number of nfsd processes on the server.
- the network performance characteristics (ethernet is assumed).
- the client configuration - the number of clients, the workload produced
by each client, and the size of client memory.
The outputs from the model will be the response time distributions of
the different NFS request types. It is anticipated that this model could
then be used to answer questions like:
What is the bottleneck in a specified system?
How many clients can a server support?
How many nfsd processes are worthwhile?
I am interested in hearing about other work in this area, including:
Similar models for any sort of file systems.
Studies of NFS performance.
Studies of NFS workload.
How to estimate cache effectiveness in a simulation model (i.e. the hit
rate) based on the number of requests and some locality estimate, rather
than on a full reference trace.
I am aware of the following things:
"File Access Performance of Diskless Workstations" by Lazowska and Zahorjan,
in TOCS, Vol 4, No 3, 1985.
Also Steve Miller (steve@umiacs.umd.edu) posted a request for information
in the area of NFS workload characterisation, and a summary to the Sun-Nets
mailing list.
I will post a summary of responses.
--
Paul Ashton
Email(internet): paul@cosc.canterbury.ac.nz
NZ Telecom: Office: +64 3 667 001 x6350
NZ Post: Dept of Computer Science
University of Canterbury, Christchurch, New Zealand