> We recently purchased several Sparc 10's to supplement our existing base
> of IPC's and Sparc 2's. We configured one of the 10's as a server, and
> upgraded all of the machines to 4.1.3. We have 200MB disks on the IPC's
> with OpenWindows, programs, and user data remotely mounted. Now, we
> seem to be getting a lot of "NFS server XXX not responding still trying"
> messages on the IPC's. I've installed patch# 100377-05 (high collision
> rate) on the server only, but it doesn't seem to help. Do I need to install
> it on all of the workstations (it says it should be a last resort).
> Any clues?
I've tried all of the suggestions given to me (see detailed suggestions
at the end of this message). They were:
1) Apply patch #10178 - Increase fd limit on inetd
2) Increase number of nfsd's on server from 8 to 16
3) Increase number of boid's on clients from 4 to 8
Alas, my problem persists.
The collision rate is low, so I don't think that is my problem.
Running etherfind(8C) and traffic(1C) provided no obvious answers. Nearly
all of my traffic is udp (nfs) traffic. Physically, it is a 10BaseT
net with 12 sun's (4 Sparc 10's, 4 IPC's, 1 Sparc 2, 1 Sparc 1+, 1 IPX,
1 3/50), 8 PC's running PC-NFS and HCL-exceed/W, and two
Macintoshes running A/UX. I have three 8 port hubs connected to each other
by 10Base2 cable (length approx 50 feet). Each 10BaseT cable is 90 feet.
At this point I'm willing to accept any WAG's and recommendations for good
(free) network analyzers.
| Randy J. Ott | randy@sisdbell.Logicon.COM |
| Logicon, Inc. | (402) 291-7750 (Voice) |
| 1408 Fort Crook Road South | (402) 291-2503 (Fax) |
| Bellevue, NE 68005 | |
> From: uunet.UU.NET!etnsed!xhaque@ucsd.EDU (Amanul Haque)
> I had the same problem. It seems that your server can not keep up with the
> requests. There is a problem with the inetd daemon of not being able to
> handle more than 40 requests per socket per second. Sun has a patch.
> I do not have the patch number, but I can find it for you if you
> need it. It might be easier for you to get it from sun. Basically
> the patch supports a command option "-r" that allows you to
> specify up to how many connections you want open per socket. I have
> all my machines running this patch (inetd -r 60 60) because all my
> machines are cross mounting each other. I also changed the MAXUSER
> variable to 16 for wach client machine and increased the # of biods
> on the server to 8 (default is 4).
> From: Jeff Mallory <access.digex.net!jeff@ucsd.EDU>
> On your SS10's, look at the number of nfsd's started up in the /etc/rc.local
> script. Factory standard is just 8, which may be insufficient for the number
> of clients you are providing for. On the server, do a ps aux | grep nfsd,
> then divide the total cpu time for a typical nfsd by the number of days since
> last reboot (from the output of the uptime command). All the nfsd's should
> have very close to the same amount of total CPU time and the average
> CPUtime/day/nfsd should be only several (~1-4) minutes for a busy, properly
> tuned system. If the CPU time for each nsfd is much larger then that, increase
> the number of nfsd's (in groups of 8, for instance) until your avg daily CPU
> time for each nfsd comes down to that area.
> Thanks for mentioning the high collision rate patch, I'm going to look that
> up and see what it applies to.
> Jeff Mallory
> From: Geert Jan de Groot <ica.philips.nl!geertj@ucsd.EDU>
> Have your network checked by an analyser. It looks like it has some problems.
> Too long perhaps? Broadcast storms? What does etherfind say?
> Geert Jan
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:07:53 CDT