Hi Alex, ... On 1/15/14 8:12 AM, "Alexander Duyck" <[email protected]> wrote:
>You mentioned using a 3.15.1 driver for the PF, I was wondering what >version of the ixgbevf driver it was you were using? 2.12.1 i.e. latest > The older versions >of the VF drivers used an older single buffer approach to receive and it >is possible that some of that 40% drop you are seeing is due to >differences in the driver instead of a difference in features. If >nothing else one thing you might try testing is the NIC passthrough >without RSC to verify how much of the benefit is the RSC versus >differences in the drivers. NIC Passthrough of a X540 NIC into a Ubuntu VM (single core) running ixgbe-3.15.1, drops about 37.5% when using only GRO vs LRO/RSC. Again - only using a few flows. > >One significant factor that can limit the performance of virtualization >is interrupt moderation. One of the effects of RSC is that the >interrupt rate can be reduced due to the fact that the number of packets >is reduced due to coalescing. Have you tried using the ethtool -C >option to configure the Rx interrupt rates so that they are the same in >both tests? I did not use ethtool but instead the module parameter InterruptThrottleRate (present both in ixgbe and ixgbevf) and set it to 1000 for all my tests. The 40% drop is with this set. Are you saying that I need to increase this value to "adaptive" ? Thanks, Chaitanya ------------------------------------------------------------------------------ CenturyLink Cloud: The Leader in Enterprise Cloud Services. Learn Why More Businesses Are Choosing CenturyLink Cloud For Critical Workloads, Development Environments & Everything In Between. Get a Quote or Start a Free Trial Today. http://pubads.g.doubleclick.net/gampad/clk?id=119420431&iu=/4140/ostg.clktrk _______________________________________________ E1000-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/e1000-devel To learn more about Intel® Ethernet, visit http://communities.intel.com/community/wired
