netdev - Re: [PATCH net-next] virtio_net: enable napi

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190614013506-mutt-send-email-mst@kernel.org>
Date:   Fri, 14 Jun 2019 02:00:10 -0400
From:   "Michael S. Tsirkin" <mst@...hat.com>
To:     Jason Wang <jasowang@...hat.com>
Cc:     Willem de Bruijn <willemdebruijn.kernel@...il.com>,
        netdev@...r.kernel.org, davem@...emloft.net,
        Willem de Bruijn <willemb@...gle.com>
Subject: Re: [PATCH net-next] virtio_net: enable napi_tx by default

On Fri, Jun 14, 2019 at 11:28:59AM +0800, Jason Wang wrote:
> 
> On 2019/6/14 上午12:24, Willem de Bruijn wrote:
> > From: Willem de Bruijn <willemb@...gle.com>
> > 
> > NAPI tx mode improves TCP behavior by enabling TCP small queues (TSQ).
> > TSQ reduces queuing ("bufferbloat") and burstiness.
> > 
> > Previous measurements have shown significant improvement for
> > TCP_STREAM style workloads. Such as those in commit 86a5df1495cc
> > ("Merge branch 'virtio-net-tx-napi'").
> > 
> > There has been uncertainty about smaller possible regressions in
> > latency due to increased reliance on tx interrupts.
> > 
> > The above results did not show that, nor did I observe this when
> > rerunning TCP_RR on Linux 5.1 this week on a pair of guests in the
> > same rack. This may be subject to other settings, notably interrupt
> > coalescing.
> > 
> > In the unlikely case of regression, we have landed a credible runtime
> > solution. Ethtool can configure it with -C tx-frames [0|1] as of
> > commit 0c465be183c7 ("virtio_net: ethtool tx napi configuration").
> > 
> > NAPI tx mode has been the default in Google Container-Optimized OS
> > (COS) for over half a year, as of release M70 in October 2018,
> > without any negative reports.
> > 
> > Link: https://marc.info/?l=linux-netdev&m=149305618416472
> > Link: https://lwn.net/Articles/507065/
> > Signed-off-by: Willem de Bruijn <willemb@...gle.com>
> > 
> > ---
> > 
> > now that we have ethtool support and real production deployment,
> > it seemed like a good time to revisit this discussion.
> 
> 
> I agree to enable it by default. Need inputs from Michael. One possible
> issue is we may get some regression on the old machine without APICV, but
> consider most modern CPU has this feature, it probably doesn't matter.
> 
> Thanks
> 

Right. If the issue does arise we can always add e.g. a feature flag
to control the default from the host.


> > 
> > ---
> >   drivers/net/virtio_net.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > index 0d4115c9e20b..4f3de0ac8b0b 100644
> > --- a/drivers/net/virtio_net.c
> > +++ b/drivers/net/virtio_net.c
> > @@ -26,7 +26,7 @@
> >   static int napi_weight = NAPI_POLL_WEIGHT;
> >   module_param(napi_weight, int, 0444);
> > -static bool csum = true, gso = true, napi_tx;
> > +static bool csum = true, gso = true, napi_tx = true;
> >   module_param(csum, bool, 0444);
> >   module_param(gso, bool, 0444);
> >   module_param(napi_tx, bool, 0644);