home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.nfs
- Path: sparky!uunet!kithrup!hoptoad!decwrl!pa.dec.com!nntpd2.cxo.dec.com!nntpd.lkg.dec.com!engage.pko.dec.com!riscee.pko.dec.com!wiryaman
- From: wiryaman@riscee.pko.dec.com (Santa Wiryaman)
- Subject: Re: Free NFS Response Time Measurement Software
- Message-ID: <1992Aug20.213642.11024@engage.pko.dec.com>
- Sender: newsdaemon@engage.pko.dec.com (USENET News Daemon)
- Organization: Digital Equipment Corporation, Maynard MA
- References: <1992Aug10.045921.13353@aim.com>
- Distribution: usa
- Date: Thu, 20 Aug 1992 21:36:42 GMT
- Lines: 42
-
- In article <1992Aug10.045921.13353@aim.com> nfsrtm@aimt.aim.com (NFS RTM Distribution) writes:
- >
- > *************************************
- > FREE NFS SERVER RESPONSE TIME UTILITY
- > *************************************
- >
- >(Note: I'm not sure if this is the proper place to post this, my
- > apologies in advance if it isn't)
- >
- >Do you want to measure the NFS response time of your servers? Have you
- >heard complaints about server performance but have lacked tools to quantify
- >the problem?
- >
- >If so, you'll want to get a copy of the FREE NFS Response Time Monitor
- >(NFSRTM) from AIM Technology. This easy-to-install-and-use command line
- >utility can be run on SPARC workstations running with the Network Interface
- >Tap (NIT) and SunOS 4.1 or higher (NOTE: the NFS servers being measured
- >DO NOT have to be running SunOS).
-
- I have been using tcpdump and awk scripts to reduce the log file. Tcpdump
- also uses NIT on Suns or packet filter on BSD UNIX and ULTRIX, I used the
- packet filter interface. What I found was insufficient using this tool was
- the 4 ms. clock resolution. On a midrange NFS server an operation such as
- getattr could easily be serviced at <1 ms. So for accuracy, a microsecond clock
- resolution was needed. What is the clock resolution of the SPARC workstations?
-
- Also, how does NFSRTM deals with timeout and retransmissions?
-
- As NFS client implementation is getting smarter, I wonder whether on the wire
- measurements are still accurate from the client perspective. The server's
- ops/sec and secs/ops are still important consideration, but as a user of the
- client I would also be concerned about the performance of the client
- implementation. So I would put more weight on performance numbers taken from
- a workload on the client. FWIW, LADDIS only measure the server performance,
- bypassing the NFS client implementation altogether.
-
-
- --
- Santa Wiryaman
-
- INTERNET : wiryaman@riscee.pko.dec.com
- UUCP : ...!decwrl!riscee.pko.dec.com!wiryaman
-