> -----Original Message----- > From: Ferruh Yigit [mailto:ferruh.yigit at intel.com] > Sent: Tuesday, November 15, 2016 2:19 PM > To: Ananyev, Konstantin; Richardson, Bruce > Cc: Vladyslav Buslov; Wu, Jingjing; Zhang, Helin; dev at dpdk.org > Subject: Re: [dpdk-dev] [PATCH] net/i40e: add additional prefetch > instructions for bulk rx > > On 10/13/2016 11:30 AM, Ananyev, Konstantin wrote: > > <...> > > >>>> > >>>> Actually I can see some valid use cases where it is beneficial to have > >>>> this > prefetch in driver. > >>>> In our sw distributor case it is trivial to just prefetch next packet on > each iteration because packets are processed one by one. > >>>> However when we move this functionality to hw by means of > >>>> RSS/vfunction/FlowDirector(our long term goal) worker threads will > >> receive > >>>> packets directly from rx queues of NIC. > >>>> First operation of worker thread is to perform bulk lookup in hash > >>>> table by destination MAC. This will cause cache miss on accessing > >> each > >>>> eth header and can't be easily mitigated in application code. > >>>> I assume it is ubiquitous use case for DPDK. > >>> > >>> Yes it is a quite common use-case. > >>> Though I many cases it is possible to reorder user code to hide (or > minimize) that data-access latency. > >>> From other side there are scenarios where this prefetch is excessive and > can cause some drop in performance. > >>> Again, as I know, none of PMDs for Intel devices prefetches packet's > data in simple (single segment) RX mode. > >>> Another thing that some people may argue then - why only one cache > >>> line is prefetched, in some use-cases might need to look at 2-nd one. > >>> > >> There is a build-time config setting for this behaviour for exactly > >> the reasons called out here - in some apps you get a benefit, in > >> others you see a perf hit. The default is "on", which makes sense for most > cases, I think. > >> From common_base: > >> > >> CONFIG_RTE_PMD_PACKET_PREFETCH=y$ > > > > Yes, but right now i40e and ixgbe non-scattered RX (both vector and scalar) > just ignore that flag. > > Though yes, might be a good thing to make them to obey that flag > properly. > > Hi Vladyslav, > > According Konstantin's comment, what do you think updating patch to do > prefetch within CONFIG_RTE_PMD_PACKET_PREFETCH ifdef? > > But since config option is enabled by default, performance concern is still > valid and needs to be investigated. > > Thanks, > ferruh
Hi Ferruh, I'll update my patch according to code review suggestions. Regards, Vlad
