SUMMARY: broken 101318-59 - not quite!

From: Paulo Licio de Geus (paulo@dcc.unicamp.br)
Date: Wed Nov 16 1994 - 22:34:01 CST


Original question:

Paulo Licio de Geus writes:
> LaCoursiere J. D. writes:
> > Fix:
> > several people suggested patches, but none knew if they worked or
> > not. Upon research of one suggested patch (101318-59), an admin
> > warned that when he installed it it broke NFS (ACK!).
>
> I've found this on my SS1000's as well. Does anyone know which
> version of 101318 is less insane regarding NFS? The machines usually
> time out on NFS mounted filesystems after a couple of days. Users get
> bored waiting for a shell to get in, if at all, till we reboot
> them. We've got $300K of pretty unusable hardware right now!
>
> Thanks in advance,
>
> --
> postmaster/manager
> Paulo Licio de Geus INTERNET: paulo@dcc.unicamp.br
> Depto de Ciencia da Computacao voice: +55 192 39-3115/8695/8442
> DCC - IMECC - UNICAMP fax: +55 192 39-7470/5808
> caixa postal: 6065
> 13081-970 Campinas SP Brazil

Summary:

101318-59 requires a bit of "seasoning" in /etc/system to work
properly on SS1000's.

Most respondents suggested going back to release 54 of the patch, plus
a lot of desperate "me-too's". However, the easiest route for me was
to try the following parameter adjustments in /etc/system, thanks to
sjk@KaPRE.COM (Scott J. Kamin):

    We saw the problems you mention when we first installed our systems
    6 months ago. We eliminated them by adjucting the values in the
    /etc/system file.

    Here's what we're running now, 4CPU, 256MB memory:

    set ufs_ninode=32768
    set ncsize=32768
    set nsegkp=1024
    set nhbuf=256
    set p_nbuf=2048
    set bufhwm=16384
    set bf:bf_nbufs=2048
    set bf:bf_nrmds1=256
    set slowscan=0x64
    set fastscan=0x2000
    set maxpgio=100

I didn't have time to even understand those settings, let alone
tweaking them to a 2CPU/128MB and a 8CPU/192MB configuration, since I
was away for most of the previous 7 days. I just used these same
values and rebooted the machines to have a go and see what happened.
A week later the machines are still up and running (thank God the
latest storms didn't cause long power failures...), no problems have
been observed since then with NFS.

Other managers might go straight to release 64, but I would suggest
having the above mods up your sleeves anyway, preferably knowing what
they do (does the manual cover all those parameters? I don't think
so...).

And hooray to the most unstable Solaris patch! Releases are updated
up to the minute:-):-).

Also mentioned:

- going back to release 54;
- checking whether automount had been stopped before applying the
patch; I saw the msg when doing it, because I installed this patch
through the 2.3.Recommended install_cluster script, which is not wise
enough. On the second installation, I believe I stopped autofs as
recommended but it made no difference. I checked
/usr/lib/autofs/automountd and it is identical to the file in the
patch. Also, the patch log file reports installation OK.
- checking whether 101344-11 (Jumbo NFS patch) was installed; This
patch is part of the 2.3.Recommended cluster and was applied at the
same time.

I apologise for the wrong "From:" field in my msg. This has already
been corrected. See my signature below for the correct address.

Thanks to:

rwh@atmos.albany.edu (Ronald W. Henderson)
tim@bullwinkle.access.com (Tim Wort)
"Frank 'Scruffy' Miller" <Frank.Miller@master.CNA.TEK.COM>
<aidan@cse.unsw.edu.au>
sjk@KaPRE.COM (Scott J. Kamin)
spillar@agric.nsw.gov.au (Ron Spillane)
<Chris_Terry/Computervision@ausns1.cv.com>
iss2jrs@uugtedc.gtedc.gte.com (John Saunders)
Casper Dik <casper@fwi.uva.nl>
boenning@igd.fhg.de (Dirk Boenning)
Tim.Griffin@fw.gs.com (Tim Griffin - Systems Consultant)
Jorge Portugal Andrade <admin@di.uminho.pt>
Anchi Zhang <anchi@Starbase.NeoSoft.COM>
Brad Burdick <bburdick@radio.com>
"Bushman, Kevin" <Bushman@comswsys.tinkernet.af.mil>
epl@Kodak.COM (Gene Loriot (epl@kodak.com))
Pug <pug@arlut.utexas.edu>
Birger.Wathne@vest.sdata.no (Birger A. Wathne)

--
postmaster/manager
Paulo Licio de Geus			INTERNET: paulo@dcc.unicamp.br
Depto de Ciencia da Computacao		voice: +55 192 39-3115/8695/8442
DCC - IMECC - UNICAMP			fax: +55 192 39-7470/5808
caixa postal: 6065
13081-970  Campinas SP Brazil



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:09:15 CDT