Rate Scaling configuration for NDIS miniport driver RRS feed

  • Question

  • I have recently encountered a usage scenario that seems not optimal from the OS side (using WindowsServer2012R2):

    I have a stress test (unrelated to networking) that uses roughly half of the available CPU cores (most of the time a subset of first N/2 cores is loaded for at least 50% and load shifts pretty randomly between these N/2 cores). The second half of CPUs (cores N/2 to N) is used much lighter (most up to ~20% and some are ~0% constantly) . Given that I invoke a network stress test (with ntttcp), expecting the OS to identify the heavier loaded CPUs and to send a new OID_GEN_RECEIVE_SCALE_PARAMETERS that will provide better balance. This doesn't happen, however - and the miniport driver remains configured with the original RSS table. Is that a shortcoming of OS load balancing algorithm? What are the triggers for the OS to send new RSS table to the miniport?

    Thursday, May 18, 2017 8:59 AM