home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!ogicse!emory!swrinde!network.ucsd.edu!cogsci!lopes
- From: lopes@cogsci.ucsd.EDU (alann lopes)
- Newsgroups: comp.os.os2.networking
- Subject: Re: IBM NFS performance (too bad!)
- Message-ID: <1854@cogsci.ucsd.EDU>
- Date: 24 Jan 93 20:36:21 GMT
- Article-I.D.: cogsci.1854
- References: <1993Jan21.180618.120415@ipgaix.unipg.it>
- Organization: University of California, San Diego
- Lines: 70
-
- In article <1993Jan21.180618.120415@ipgaix.unipg.it> luca@dedalo.unipg.it (Luca Priorelli) writes:
- -Every once in a while I get upset of the poor performance of
- -IBM NFS and hope that new CSDs would solve the problem.
- -Instead... no way.
- -
- -I have just applied the latest CSD found in hobbes and:
- -
- -- I can now mount an exported OS/2 directory from a SPARCstation
- - (before I couldn't) but when I try to copy a file from the
- - Sun to the OS/2 (running on a PS/2 95), well... it takes about 11 minutes
- - to copy a 2Mbytes file!!!
- -
- -- Emacs does not work yet.
- -
- -And here are some timings copying a 2Mbytes file:
- -
- -machine type
- -A ps/2 40 386sx 20Mhz
- -B ps/2 80 386dx 20Mhz
- -C ps/2 95 486dx 33Mhz
- -D SPARCstation 33Mhz
- -
- -A --> C (write) 39"
- -A <-- C (read) 49" (this is the first time a see a NFS read taking
- - more than a write)
- -
- -A --> D 35"
- -A <-- D 1'05" (??? more than above!)
- -
- -B --> D 35"
- -B <-- D 20" (well, sounds better!)
- -
- -C --> D 26"
- -C <-- D 6" (not bad)
- -
- -D --> C 11'10" (??? this must be a BIG BUG)
- -D <-- C 2" (I cannot believe this!)
- -
- -Well, can you find any sense from these figures?
- -
- -Why D reads in 6" and A in 1'05"? OK it is a faster machine but this cannot
- -explain the difference. Moreover, it takes more to read than to write!
- -B which is almost the same speed does much better.
- -
- -The ethernet adapter? 3C509 for A and 3C525 for B and C. 3C509 is
- -advertized as a very fast one (or it isn't?). The counterproof is
- -easy. I have installed LAN server 3.0 on C and LAN requester on A.
- -Well, I cannot tell the difference between a local disk and the remote
- -disk from C.
- -
- -Why IBM NFS is so bad?
- -
- -Luca Priorelli
- -
-
- I wish I knew! I've been told by their tech support that they're
- rewriting a lot of the NFS code, but this was about 2 months ago.
-
- I'm happy with the base tcp/ip package, although there are quarks
- in the terminal support (ansi, vt100, etc...), but NFS is horrible.
- I've been complaining about this since Sept 92 and they appeared to
- be aware of the problem back then. We're approaching 5 months and I'm
- still getting the same bogus performance ( 20-40K /sec average
- read/write ). This is with a 7168 buffer, it's much worse at 4096.
- What kind of read/write transfer rates are people getting????
-
- Thanks -- alann
-
- alann lopes: alopes@ucsd.edu -- internet
- (619) 534-5438 ALOPES@UCSD -- bitnet
-