[Cake] Peeling GRO

Eric Dumazet edumazet at google.com
Tue Apr 28 18:16:40 EDT 2015


Adding a limit on number of segments aggregated by GRO is doable.

We have such a patch at Google, mainly because we run a 3.11 based kernel,
and we increased MAX_SKB_FRAGS from 17 to 45

Doing so, we added a sysctl to be able to quickly 'revert' this increase in
case it had some catastrophic consequence.

cat /proc/sys/net/core/gro_max_frags
45



On Tue, Apr 28, 2015 at 3:12 PM, Dave Taht <dave.taht at gmail.com> wrote:

> On Tue, Apr 28, 2015 at 3:04 PM, Eric Dumazet <edumazet at google.com> wrote:
> > This what we do in net/sched/sch_tbf.c  (search for tbf_segment())
>
> Thank you! I had no idea that what I had been calling peeling already
> existed. Perhaps in the simplest.qos model in the sqm-scripts (only
> one sub qdisc) we should try tbf instead of htb?
>
> > Although this is quite expensive, this is doable.
>
> Turning off GRO universally is also expensive (in terms of costs to
> other devices that are not rate limited, and in terms of actually
> getting it right in the first place)
>
> the mvneta dropped from 720Mbit each way to 320 with offloads off. We
> have had similar trouble with the ubnt edgerouters (8 ports worth of
> offloads).
>
> We were mostly struggling with stuff
> at very low rates < 60Mbit, now we are struggling with 100-300Mbit
> inbound....
>
> >
> >
> >
> > On Tue, Apr 28, 2015 at 2:38 PM, Dave Taht <dave.taht at gmail.com> wrote:
> >>
> >> Dear Eric:
> >>
> >> how would you go about peeling apart a TSO/GSO/GRO packet in a qdisc,
> >> and then serving up one packet (or drr quantum's worth) of packets?
> >>
> >> I am seeing 64K GRO'd packets from mvneta driver in the linksys 1900ac
> >> device in particular.
> >>
> >> ---------- Forwarded message ----------
> >> From: Dave Taht <dave.taht at gmail.com>
> >> Date: Tue, Apr 28, 2015 at 11:04 AM
> >> Subject: documentation review request and out of tree cake builds for
> >> openwrt/etc.
> >> To: cake at lists.bufferbloat.net, bloat <bloat at lists.bufferbloat.net>,
> >> "cerowrt-devel at lists.bufferbloat.net"
> >> <cerowrt-devel at lists.bufferbloat.net>
> >>
> >>
> >> 1) I have burned much of the morning updating this:
> >>
> >> http://www.bufferbloat.net/projects/codel/wiki/Cake
> >>
> >> Reviewers wanted.
> >>
> >> 2) I am not sure of the right configuration to push stuff to the
> >> iproute2-cake repo (removing sfq_codel, and there is an important
> >> bugfix in the mainline iproute2 that needs to be pushed into it)
> >>
> >> 3) I have pushed support for building cake out of tree in openwrt to
> >> the ceropackages repo (for now), but have not created a new
> >> iproute2-cake for it due to item 2. My intent would be to get both
> >> into openwrt chaos calmer for further testing... in some repo or
> >> another that wasn't ceropackages. the routing repo?
> >>
> >> 4) http://www.dslreports.com/forum/news,133536~fmode=flat~days=2000
> >> has some good convo going on also.
> >>
> >> --
> >> Dave Täht
> >> Open Networking needs **Open Source Hardware**
> >>
> >> https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67
> >>
> >>
> >> --
> >> Dave Täht
> >> Open Networking needs **Open Source Hardware**
> >>
> >> https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67
> >
> >
>
>
>
> --
> Dave Täht
> Open Networking needs **Open Source Hardware**
>
> https://plus.google.com/u/0/+EricRaymond/posts/JqxCe2pFr67
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/cake/attachments/20150428/4980a80f/attachment-0002.html>


More information about the Cake mailing list