Comware

 View Only
  • 1.  Procurve 2524 switch and PCM 2.2

    Posted Sep 03, 2007 12:01 AM
    hello,
    we use PCM+ 2.2 and nagios

    When "HP ProCurve Traffic Launch Service" is started, the CPU load of Procurve 2524 switches stay above 50% and sometime the switch doesn't respond to a ping command from Nagios.

    if I stop "HP ProCurve Traffic Launch Service" the CPU of the switch decrease to 16/17%.
    are there some parameters on PCM 2.2 for old switches ? to reduce the CPU load .

    regards

    Pascal


  • 2.  RE: Procurve 2524 switch and PCM 2.2

    Posted Sep 03, 2007 01:56 AM
    It's most likely due to XRMON which can be a bit resource intensive. On those switches maybe it's better to use 'statistics' only. Disable 'sampling' on them.

    Alternatively, limit the ports to which sampling is enabled, possibly use it only on the uplink port.


  • 3.  RE: Procurve 2524 switch and PCM 2.2

    Posted Sep 03, 2007 05:20 AM
    thank you,
    I will try

    Pascal


  • 4.  RE: Procurve 2524 switch and PCM 2.2

    Posted Sep 07, 2007 04:26 AM
    Hi,

    I have removed statistics using "Manually disable sampling and statistics" on all ports exept the uplink but it doen't solve the problem.

    are there other parameters to reduce the load of the switch ?

    regards

    Pascal


  • 5.  RE: Procurve 2524 switch and PCM 2.2

    Posted Sep 07, 2007 05:05 AM
    If you also remove it from the uplink does that drop the CPU down enough? In my testing by sampling only one port it would go up to 40% or so.

    If you just use 'statistics' only on the switch ports that should be okay, you just won't see the extra detail that 'sampling' gives you (tcp source and destination port, etc).

    Regardless of the CPU utilisation, it should not affect layer-2 performance of the switch. Only the management will be a bit slower. If you find this is causing problems with Nagios, possibly set it to more retries.


  • 6.  RE: Procurve 2524 switch and PCM 2.2

    Posted Sep 11, 2007 05:08 AM
    Hello,

    I have done some tests

    On a 2524 switch
    without sampling & stats : CPU Util = 17-19%
    with stats on 1 port : CPU Util = 17-20%
    with sampling & stats on 1 port : CPU Util = 30-32%

    without sampling & stats : CPU Util = 17-19%
    with stats on 2 ports : CPU Util = 17-25%
    with sampling & stats on 2 ports : CPU Util = 33-37%

    I try to use only statistics on 1 or 2 ports.
    thank you for your help

    Pascal