Re: Diagnosing network slowness: server or network?

From: Walter Roberson (roberson_at_ibd.nrc-cnrc.gc.ca)
Date: 06/24/05

  • Next message: Doug Freyburger: "Re: Diagnosing network slowness: server or network?" 
    Date: Fri, 24 Jun 2005 15:44:28 +0000 (UTC)

    In article <bqVue.295$Tk2.256@trnddc02>,
    sinister <sinister@nospam.invalid> wrote:
    :The curious thing is that all the cheap Suns with *really* bad problems
    :(i.e., graphics taking minutes to load, not just ftp slowness) show no
    :intervening hosts/switches/whatever when I run traceroute. Hosts that show
    :ftp slowness but graphics relatively unaffected show intermediate switches
    :according to traceroute.

    switches don't show up in traceroute -- only layer 3 hops decrement
    the TTL field so only layer 3 hops can return icmp ttl exceeded messages.

    Your v1280 server is unlikely to have more than a couple of ports,
    so you almost certainly have a layer 2 switch in the middle that is
    not showing up.

    Anyhow, your description sounds very much like duplex problems to me.
    I would suggest that you should launch right in to forcing the
    speed and duplex on the v1280 server and the switchport it is
    connected to: instances in which forcing speed and duplex on
    both ends make things -worse- are quite uncommon (but not unknown :( )

    The test after that would be to force speed and duplex on one of
    the cheap Suns you mention (and corresponding switchport).

    There is no firm agreement about autonegotiation vs forced
    speed and duplex. The rule of thumb that seems to be most common
    is that speed and duplex should be fixed for critical infrastructure
    (switches, routers, key servers), but that autonegotiation should be
    used for desktops. (Opinions vary on this, though!!) The practical
    idea behind this particular rule of thumb is that you can't afford
    to have speed and duplex issues on the key infrastructure that does
    not change very often, but that desktops tend to change too often
    to make it -convenient- to lock speed and duplex for all of them...
    and if something does go wrong for the desktops, it isn't going
    to affect very many people. 

    -- 
   Beware of bugs in the above code; I have only proved it correct,
   not tried it.                             -- Donald Knuth
  • Next message: Doug Freyburger: "Re: Diagnosing network slowness: server or network?"

    Relevant Pages

    • Re: Duplex/Speed Hardcoding
      ... Have a colleague who insists every port on every switch be hardcoded to ... general app - that they now only hardcode the duplex settings? ... A long while ago we had a pair of non-Cisco switches and if you ...
      (comp.dcom.sys.cisco)
    • Re: How to safely remove an external SCSI device?
      ... duplex of a network interface under solaris. ... It's not switches that are wrong, it's always sun sparc based hardware and the intense desire for 100/half and other stupid stuff like that. ...
      (comp.sys.sun.admin)
    • Re: Diagnosing network slowness: server or network?
      ... >:ftp slowness but graphics relatively unaffected show intermediate switches ... > speed and duplex on the v1280 server and the switchport it is ... mii-tool or/and eth-tool, allowing to check/set settings on the ...
      (comp.unix.admin)
    • Re: Poor network performance: a lot of timeouts
      ... I had the nics running 100baseTX Full Duplex. ... >Changing this to autosense made the problems gone. ... (on most unmanaged switches), fail, and will fall back to half duplex ...
      (freebsd-stable)
    • Re: 1 PC cant find domain controller
      ... setting is set to something that is causing the problem. ... ports on the 3Com switches were set to Full Duplex and it was set to Half ...
      (microsoft.public.windows.server.networking)