On Fri, Nov 5, 2010 at 12:20 PM, Claudio Jeker <[email protected]> wrote: > The second generation ix(4) (82599) produce a low latency interrupt when > the RX DMA ring gets full. This interrupt is currently unmoderated. On > overload mclgeti() shrinks the ring to very small sizes and ix(4) will > answer that with a massive interrupt storm. Enabling interrupt moderation > on the LLI is possible and we should enable it since we do not use the > LLI for any other low latency reason. > > With this diff the throughput of my test system increased from under > 200kpps to almost 500kpps when overloading the system with 1Mpps on the > input side. > > Please test on 82598 and 82599. > -- > :wq Claudio >
i'm okay with the diff (:
