Re: advice on tuning sol8 nfs servers

From: CJT (abujlehc_at_prodigy.net)
Date: 07/07/05

  • Next message: dare_eagle_at_yahoo.com: "Telnet without swap space"
    Date: Thu, 07 Jul 2005 04:03:14 GMT
    
    

    Steven C. Liu wrote:
    > Hi,
    >
    > I work for a very large US financial services firm, and lately we've
    > noticed a huge performance hit when accessing our NFS volumes. Our log
    > files show copious "NFS server not responding" messages. This is very
    > worrisome when the financial markets are open. Any advice on how to
    > maximize NFS performance is appreciated.
    >
    > Some info on our environment:
    > 1. We use VCS to manage the disk volumes as well as the server-side NFS
    > processes.
    > 2. Our clients mount from the NFS servers using the logical names, not
    > the physical - i.e., logical, for VCS failover for HA.
    > 3. Our disk volumes are huge - e.g., a single 1.5 TB disk group (from
    > too many 25 GB LUNs from Sym EMC storage) divided into 3 logical volumes.
    > 4. Our NFS servers are Sol8, but we can use a different OS/platform if
    > there is a convincing argument for it. The h/w are real server-class
    > platforms, not headless desktops.
    > 5. Our clients are predominantly Linux (RHES3) on powerful HP PC's; we
    > also see problems with our Sol2.6/Sol8/Sol9 clients (running on modern
    > h/w platforms).
    >
    > Suggestions can be anything relevent: kernel params, mount settings, #
    > of nfsd procs, volume size, OS, hardware platform, # of clients per
    > server, etc. Anything.
    >
    > TIA,
    >
    > Steve

    Some thoughts ...

    Are you sure it's the servers and not a bottleneck in the network?
    Pretty much the only time I see those messages is when I have a link
    that's failed/failing. I think the clients try for quite a while
    before giving up.

    What do your CPU and I/O utilization numbers on the server(s) look like?
    Do the machines have enough memory?

     From your description it sounds like the problem didn't always exist --
    what has changed?

    I think Solaris 9 made some significant improvements over 8, but you'd
    need to do a careful study of your particular situation before shifting.
    Solaris 10 promises still more enhancements, but is still relatively
    new.

    -- 
    The e-mail address in our reply-to line is reversed in an attempt to
    minimize spam.  Our true address is of the form che...@prodigy.net.
    

  • Next message: dare_eagle_at_yahoo.com: "Telnet without swap space"

    Relevant Pages

    • Re: advice on tuning sol8 nfs servers
      ... > I work for a very large US financial services firm, ... > noticed a huge performance hit when accessing our NFS volumes. ... Are you Linux clients mounting UDP or TCP? ...
      (comp.sys.sun.admin)
    • Re: NFS- SAN - FreeBSD
      ... The way to use a DAS is to connect the DAS to a server with an external SAS cable. ... You are quite right, I would like to use NFS to connect the device to the 6 servers I have, again, it would be only hosting the /home partition for each of them. ... Would freebsd support (on the storage device) that many connections? ...
      (freebsd-questions)
    • Re: Stale NFS file handles on 8.x amd64
      ... minimum of NFS problems, but it got worse with 8.x. ... accessing mail on a Netapp over NFSv3 via imapd. ... Client connections to imapd go to random servers ... I upgraded some of the servers to 8.x and dovecot 1.2 and ran ...
      (freebsd-questions)
    • Re: NFS mounts to NetApp
      ... is to gather some statistics from your Sun servers to verify throughput ... Netapp's through our Sun server we get something like: ... > having regarding NFS timeouts to NetApp filers from Sun Solaris boxes. ... > We have a NetApp filer cluster where each partner has a fiber gig ...
      (comp.unix.solaris)
    • Re: The dreaded "Alternatives to NFS" question
      ... >> to server secure data that is stored on servers based in a DMZ. ... > over NFS (unless you consider the software itself so confidential that ... confidential data between hosts but you're still wrong. ...
      (comp.security.unix)