Re: [PATCH] net/i40e: add packet prefetch

[Vladyslav Buslov <Vladyslav.Buslov@harmonicinc.com>]

Ferruh,

In our case patch significantly improves application performance. (~40% more PPS on load balancer core)
Using DPDK examples I can only reproduce perf improvements with similar design apps like load_balancer.
With applications that send on free packets on same core where they are received performance is not improved.

It seems that this patch doesn't do anything(or even actually slightly degrades performance) for run-to-completion apps so we will have to continue maintaining it as part part of our internal branch.

Regards,
Vlad

> -----Original Message-----
> From: Ferruh Yigit [mailto:ferruh.yigit@intel.com]
> Sent: Monday, April 03, 2017 1:31 PM
> To: Pei, Yulong; Vladyslav Buslov; Zhang, Helin; Wu, Jingjing
> Cc: dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH] net/i40e: add packet prefetch
> 
> On 4/1/2017 3:01 AM, Pei, Yulong wrote:
> > Hi All
> >
> > In Non-vector mode, without this patch, single core performance can
> > reach 37.576Mpps with 64Byte packet, But after applied this patch , single
> core performance downgrade to 34.343Mpps with 64Byte packet.
> 
> Thanks Yulong for the test.
> 
> Vladyslav,
> 
> Is above test results match with your result?
> 
> If there is a degradation, this patch should not be merged.
> 
> Thanks,
> ferruh
> 
> >
> > Best Regards
> > Yulong Pei
> >
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Vladyslav Buslov
> > Sent: Wednesday, March 1, 2017 6:57 PM
> > To: Zhang, Helin <helin.zhang@intel.com>; Wu, Jingjing
> > <jingjing.wu@intel.com>; Yigit, Ferruh <ferruh.yigit@intel.com>
> > Cc: dev@dpdk.org
> > Subject: [dpdk-dev] [PATCH] net/i40e: add packet prefetch
> >
> > Prefetch both cache lines of mbuf and first cache line of payload if
> CONFIG_RTE_PMD_PACKET_PREFETCH is set.
> >
> > Signed-off-by: Vladyslav Buslov <vladyslav.buslov@harmonicinc.com>
> <..>