home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.nfs
- Path: sparky!uunet!sun-barr!cs.utexas.edu!sdd.hp.com!mips!odin!fido!zola!twilight!speaker.wpd.sgi.com!coolidge
- From: coolidge@speaker.wpd.sgi.com (Don Coolidge)
- Subject: Re: NFS I/O Ops/seconds
- Message-ID: <nm0e0n8@twilight.wpd.sgi.com>
- Sender: news@twilight.wpd.sgi.com ( CNews Account at twilight.wpd.sgi.com )
- Organization: Silicon Graphics, Inc.
- References: <1992Jul22.061146.15641@u.washington.edu> <l6r4uvINNf0p@exodus.Eng.Sun.COM> <nkok0p0@rhyolite.wpd.sgi.com> <1992Jul23.192203.12376@newshost.lanl.gov>
- Date: Thu, 23 Jul 92 23:13:05 GMT
- Lines: 45
-
- In article <1992Jul23.192203.12376@newshost.lanl.gov>, cwm@beta.lanl.gov (Christina W Mercier) writes:
- |> In article <nkok0p0@rhyolite.wpd.sgi.com> vjs@rhyolite.wpd.sgi.com (Vernon Schryver) writes:
- |> >In article <l6r4uvINNf0p@exodus.Eng.Sun.COM>, beepy@tabitha.Eng.Sun.COM (Brian Pawlowski) writes:
- |> >>
- |> >> The number "300 NFS ops/s per ether" is a rule-of-thumb
- |> >>
- |> >Yes, everyone keeps saying, for ethernet.
- |> >One can hope to do significantly better on FDDI.
- |>
- |> So does anyone have a start at a rule-of-thumb for FDDI????
- |> Any results of default-mix NHFSSTONE benchmarks on FDDI would be
- |> appreciated.
-
- There's really no rule-of-thumb, because as yet nobody hits FDDI
- bandwidth for any real applications (OK, maybe Cray excluded :^).
- However...
-
- Let's say your FDDI ttcp throughput can hit 6 Mbytes/sec (that's
- less than 50% bandwidth, but very few FDDI implementations are
- that fast yet). That's improvement over Ethernet throughput by
- a factor of 6 (most boxes can saturate an Ethernet these days).
- Since only roughly 1/3 of the nhfsstones default mix does reads
- or writes, they are the only operations affected by medium bandwidth
- (the others are basically latency-limited, which is about the same
- for Ethernet as for FDDI - speed of light, ya know...). So, your reads
- and writes would complete in 1/6 the time, and everything else would
- be unchanged. Net result - the overall FDDI number is less than
- 1.4 times the Ethernet number.
-
- If your FDDI ran at a full 100Mb/sec (12.5 MB), the best you could
- hope for would be improvement by a bit over a factor of 1.4.
-
- Remember, FDDI has no effect (actually very little) on the IOPS
- for all latency-bound operations, only for reads and writes. So
- even if you had infinite bandwidth, you'd never get a larger
- improvement factor than 1.5 .
-
- Somebody should have pointed this out to the guys specifying LADDIS;
- their recommendations for adding resources (extra networks, for
- instance) seem to think that because FDDI bandwidth is 10X Ethernet,
- that means you'll get 10X your Ethernet nhfsstones numbers. Well,
- it just ain't so.
-
- Don Coolidge
- coolidge@speaker.wpd.sgi.com
-