DX NetOps

Expand all | Collapse all

Spectrum 10: Performance experience regarding number of devices/models

  • 1.  Spectrum 10: Performance experience regarding number of devices/models

    Posted 07-09-2015 05:39 AM

    Hello

     

    According to the release notes of Spectrum 10 we see huge improvements regarding number of devices/models per server (due to 64-bit support).

     

    Has anyone already tested this new scalability and can share their findings/results?

     

     

    Kind regards,

      Leo Bär



  • 2.  Re: Spectrum 10: Performance experience regarding number of devices/models

    Broadcom Employee
    Posted 07-16-2015 08:20 AM

    Hi Leo_Baer,

     

    Sharing one of the thread on Spectrum ideas community: Speed up the  model activation process in Spectrum.

     

    Community members talking about improvement in model activation with Spectrum 10.0.

     

    Thanks,

    Nagesh



  • 3.  Re: Spectrum 10: Performance experience regarding number of devices/models

    Broadcom Employee
    Posted 07-16-2015 10:29 AM

    Hello Leo,

     

    overall the CA Spectrum R10 64bit improvements will help to allow more "concurrent threading" for the SpectroSERVER - and to load more data "into the stack". But this is a valid improvement in case of a "static" network and without a detailed view to the modeling. Modeling is one factor which can affect the internal CA Spectrum logic - like the Fault-Isolation and the RootCauseAnalysis.

     

    Therefore - knowing the current CA Spectrum runtime - i.e. on R9.3 or R9.4 flavours - needs to be reviewed. Main point to start is the SSPerformance application model which is available per single Landscape /SpectroSERVER. Have a look to the SSPerformance application model - find these via Locater Search - check for the Component Details - Performance tab -- and here select (i.e. for 6 hours):

     

    SS_Idle_Time -- 100% indicates any new thread can immediatelly start / 0% means - no more thread right now (hanging thread / single thread-condition)

    SS_Timer_Latency  -- which is delay in seconds

    SS_Polling_Latency -- which is delay in seconds

     

    In case of temporary high workload condition the "SS_Idle_Time" may show for a 1 minute 0%_condition - which affects anything at SpectroSERVER level - then you may find polling and timer latencies. Having a high loaded SpectroSERVER may require to re-poll many devices - much more then before at R9.4 workload. So - in general there is no common rule that we can suggest to double the number of devices - or more for a SpectroSERVER with Release 10 (versa current R9.3/R9.4). Knowing the current CA Spectrum usage and workload (i.e also the events per second / alarms per second / average) is important.

     

    Next then is - to have on top of a most_current_hardware SpectroSERVER - running a OC-server with also high performance platform being able to process and propagate the high workload.

    In general it is true - the R10 SpectroSERVER allows much higher model load - we tested with 800k models and is runs with good performance. But again - please honor the fact of this 3 tier application design and it is important to have SpectroSERVER and OC-server running on modern platform. Cheers, Joerg