home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.sys.sun.hardware
- Path: sparky!uunet!mcsun!sunic!ericom!exucom.exu.ericsson.se!news
- From: exudnw@exu.ericsson.se (Dave Williams)
- Subject: Re: NFS I/O Ops/seconds
- Message-ID: <1992Jul22.172312.12965@exu.ericsson.se>
- Sender: news@exu.ericsson.se
- Nntp-Posting-Host: floyd.exu.ericsson.se
- Reply-To: exudnw@exu.ericsson.se
- Organization: Ericsson Network Systems
- References: <1992Jul22.061146.15641@u.washington.edu>
- Date: Wed, 22 Jul 1992 17:23:12 GMT
- Lines: 51
-
- In article 15641@u.washington.edu, kint@rio.engr.washington.edu (Rick Kint) writes:
- > We're buying a Sun to serve as an NFS server in one of our departments.
- >The Sun rep commented that a typical Ethernet allows 300 NFS ops per second,
- >so if you're on a single wire any server bandwidth beyond that is wasted.
- >
- > I have no feel whatsoever for the numbers that go into this figure,
- >can anyone in netland (a) explain this in words of two syllables or less, and
- >(b) does anyone know the figures for current systems?
- >
- > I'm sure that this includes assumptions about what represents a
- >typical mix of operations; since a getattr is presumably quicker than a write
- >(for example), how useful are these numbers anyway?
- >
- > ObCurmudgeonlyRemark: I was reading about the SS10/41, with 1 MB of
- >cache. Good Lord, a meg of cache on a workstation, why when I was your age
- >we thought a meg of *main memory* on a *mainframe* was a lot...
-
- I did some testing a while back and found that you can saturate a network
- segment with 2 sun IPCs at around 320 NFSops with the "Legato mix" and
- Nhfsstone. This is a pathological test, though, so we developed some
- rules based on trial/error.
-
- Rules of thumb we use:
-
- 1) For diskless (Sun) clients, no more than 20 clients per subnet.
-
- 2) Use network co-processors (NC400/600) whenever possible. You can get
- up to 60 diskless clients on a 490 with them (remember rule 1).
-
- 3) Dataless clients use *much* less bandwidth (no root, swap, usr, etc.)
- over the net. Your results will depend more on what the users data
- requirements are than on anything else. If you are running [e,m]cad
- or huge databases you will be network-bound. Small disks are now
- reliable and cheap - consider this seriously.
-
- 4) Get the fastest server drives you can afford (we use IPI, but new fast
- SCSI might be ok).
-
- 5) Never, never run user jobs on the server or allow users to have
- interactive logins on the server. NFS is shy - it just wants to be
- left alone...
-
- 6) Prestoserve write accelerators only benefit systems with write % above
- 15% (ours are around 6%).
-
- ---
- = exudnw@exu.ericsson.se =
- = David Williams "You can't win, you can't break even, =
- = Ericsson Network Systems and you can't quit" =
- = Richardson, TX 75081 my opinions... =
-
-