[prev in list] [next in list] [prev in thread] [next in thread] 

List:       tru64-unix-managers
Subject:    SUMMARY: System Crawling
From:       "<cceckw () leonis ! nus ! sg>" <cceckw () leonis ! nus ! sg>
Date:       1995-08-31 4:49:53
[Download RAW message or body]

Dear OSF Managers,

I only received 1 reply so far and it came from Hellebo Kgut.  Well 
thanks to you Mr. Hellebo Kgut.

It did give us something to think about but we won't have the chance to 
implement it since we'll be re-configuring the system.  The reply from 
DEC engineering was that the system was overloaded (we had more  
than 400 concurrent users accessing the file server at that time).  The 
recommendation was to move the file systems back to the login machine 
where there are 400 users in order to cut down on nfs traffic.  As they 
pointed, NFS is only good for a bunch of workstations and certainly not 
for 400 users.  


Regards,
Chang

-----------------------------------------------------------------------

The original question was:
Dear osf managers,

We're running OSF/1 v3.2a on an alphaserver 2100 4/200 as an NFS file 
server.  This being the case, it usually has only 1 or 2 users on the 
system.  It also serves as a mail server.  As expected, the load from the 
uptime command was 1 to 2.  The number of users accessing the file server 
was about 400.

Everything went well until one day, the load went up to 20.  The output 
from the ps aux command shows kernel idle process chewing up 90 over 
percent of the CPU.  At that time, there were no other processes running 
since we stopped sendmail.  This is certainly not normal.  The dec 
engineers then changed the FDDI card but this didn't help.  Has anyone of 
you experience this?

A few days ago, we added another CPU to the system but found that cpu 0 
was always very busy (about 90% busy) whereas cpu 1 was very idle (about 
10% busy).  A dec engineer, after seeing the output of kdbx (proc and 
thread) commented that it was the nfs threads that were chewing up the 
cpu.  If so, is it possible to spread the load evenly across the 2 cpus?

The system is now c r a w l i n g and users get stuck in their pine for 
something like 5 to 10 minutes. 

I'd appreciate any comments or suggestions.

Thank you all for your assistance.

Regards,
Chang

The only reply was:

From: Hellebo Knut <bgk1142@bggfu2.nho.hydro.com>

Have you consulted the 'System Tuning and performance guide' ?
There is a nice chapter about NFS and such. Try to increase the number of
nfsd daemons to one of the numbers suggested in the manual (our NFS server
runs 32 nfsds)

-- 

      ******************************************************************
      *         Knut Helleboe                    | DAMN GOOD COFFEE !! *
      *         Norsk Hydro a.s                  | (and hot too)       *
      * Phone: +47 55 996870, Fax: +47 55 996342 |                     *
      * Pager: +47 96 500718                     |                     *
      * E-mail: Knut.Hellebo@nho.hydro.com       | Dale Cooper, FBI    *
      ******************************************************************


===============================================================================
  Chang Koon Wang
  National University of Singapore
  Computer Centre				     Tel: 772-2477
  10, Kent Ridge Crescent			     Fax: 778-0198
  Singapore 0511				Internet: cceckw@nus.sg
===============================================================================

[prev in list] [next in list] [prev in thread] [next in thread]