> -----Original Message----- > From: Maxime Coquelin [mailto:maxime.coquelin at redhat.com] > Sent: Monday, August 22, 2016 6:02 PM > To: Wang, Zhihong <zhihong.wang at intel.com>; dev at dpdk.org > Cc: yuanhan.liu at linux.intel.com > Subject: Re: [PATCH v3 0/5] vhost: optimize enqueue > > > On 08/22/2016 10:11 AM, Maxime Coquelin wrote: > > Hi Zhihong, > > > > On 08/19/2016 07:43 AM, Zhihong Wang wrote: > > > This patch set optimizes the vhost enqueue function. > > > > > > It implements the vhost logic from scratch into a single function > > > designed > > > for high performance and good maintainability, and improves CPU > > > efficiency > > > significantly by optimizing cache access, which means: > > > > > > * For fast frontends (eg. DPDK virtio pmd), higher performance > (maximum > > > throughput) can be achieved. > > > > > > * For slow frontends (eg. kernel virtio-net), better scalability can be > > > achieved, each vhost core can support more connections since it takes > > > less cycles to handle each single frontend. > > > > > > The main optimization techniques are: > > > > > > 1. Reorder code to reduce CPU pipeline stall cycles. > > > > > > 2. Batch update the used ring for better efficiency. > > > > > > 3. Prefetch descriptor to hide cache latency. > > > > > > 4. Remove useless volatile attribute to allow compiler optimization. > > > > Thanks for these details, this is helpful to understand where the perf > > gain comes from. > > I would suggest to add these information as comments in the code > > where/if it makes sense. If more a general comment, at least add it in > > the commit message of the patch introducing it. > > Indeed, adding it to the cover letter is fine, but the information is > > lost as soon as the series is applied. > > > > You don't mention any figures, so I set up a benchmark on my side to > > evaluate your series. It indeed shows an interesting performance gain. > > > > My setup consists of one host running a guest. > > The guest generates as much 64bytes packets as possible using > > pktgen-dpdk. The hosts forwards received packets back to the guest > > using testpmd on vhost pmd interface. Guest's vCPUs are pinned to > > physical CPUs. > > > > I tested it with and without your v1 patch, with and without > > rx-mergeable feature turned ON. > > Results are the average of 8 runs of 60 seconds: > > > > Rx-Mergeable ON : 7.72Mpps > > Rx-Mergeable ON + "vhost: optimize enqueue" v1: 9.19Mpps > > Rx-Mergeable OFF: 10.52Mpps > > Rx-Mergeable OFF + "vhost: optimize enqueue" v1: 10.60Mpps > > > I forgot to add that before this series, I think we should first fix the > windows bug. > Else we will need a dedicated fix for the stable branch.
Okay I'll try to fix it, though I can't make any promises at present. Have tried once but stopped since we don't have enough debug info from the frontend side so basically I was debugging the backend based on guesses. > > Regards, > Maxime
