none
Server consolidation results don't make sense given observed performance metrics RRS feed

  • Question

  • MAP 9.3 reports performance metrics for a physical server with two E5-2620 CPUs (6 cores each @ 2.0 ghz) has average CPU Utilization of 56.1%, Max 61.46%, and 95th percentile 57.3%.

    When I run the Server Consolidation wizard using 75% resource utilization targets, I get a "processor utilization of guest exceeds CPU capacity available on host" error in the "Last placement failed reason" column of the ServerVirtRecommendations report when I simulate putting just this one workload on a host with any of these CPUs:

    * Two E5-2620 (exact same as the current physical server)

    * Two E5-2637 v2 (4 cores each @ 3.5 ghz)

    * Two E5-2690 (8 cores each @ 2.9 ghz)

    * Two E5-2690 v2 (10 cores each @ 3 ghz)

    It WILL place on this one:

    * Two E5-2637 v3 (4 cores each @ 3.5 ghz)

    So... haw can a workload that needs 60% of an 8-core 2ghz server NOT run well on a server using 20 cores of the next CPU generation each running 50% faster? And if the load can't be handled by that 20-core server, how can an 8-core 3.5ghz v3 host handle it? I know v3 is better, but the jump was not THAT huge. 

    What gives?

    Tuesday, May 10, 2016 9:20 PM