[lkml]   [2009]   [Dec]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v2 0/4] Defer skb allocation for both mergeable buffers and big packets in virtio_net
On Sun, 2009-12-13 at 12:19 +0200, Michael S. Tsirkin wrote:
> Shirley, some advice on packaging patches
> that I hope will be helpful:
> You did try to split up the patch logically,
> and it's better than a single huge patch, but it
> can be better. For example, you add static functions
> in one patch and use them in another patch,
> this works well for new APIs which are documented
> so you can understand from the documentation
> what function should do, but not for internal, static functions:
> one ends up looking at usage, going back to implementation,
> back to usage, each time switching between patches.
> One idea on how to split up the patch set better:
> - add new "destroy" API and supply documentation
> - switch current implementation over to use destroy API
> - split current implementation into subfunctions
> handling mergeable/big cases
> - convert functions to use deferred allocation
> [would be nice to convert mergeable/big separately,
> but I am not sure this is possible while keeping
> patchset bisectable]
> Some suggestions on formatting:
> - keep patch names short, and prefix with module name,
> not patchset name, so that git summaries look nicer. E.g.
> Defer skb allocation -- add destroy buffers function for virtio
> Would be better "virtio: add destroy buffers function".
> - please supply commit message with some explanation
> and motivation that will help someone looking
> at git history, without background from mailing list.
> E.g.
> "Add "destroy" vq API that returns all posted
> buffers back to caller. This makes it possible
> to avoid tracking buffers in callers to free
> them on vq teardown. Will be used by deferred
> skb allocation patch.".
> - Use "---" to separate description from text,
> and generally make patch acceptable to git am.
> It is a good idea to use git to generate patches,
> for example with git format-patch.
> I usually use it with --numbered --thread --cover-letter.
> > Guest virtio_net receives packets from its pre-allocated vring
> > buffers, then it delivers these packets to upper layer protocols
> > as skb buffs. So it's not necessary to pre-allocate skb for each
> > mergable buffer, then frees it when it's useless.
> > This patch has deferred skb allocation when receiving packets for
> > both big packets and mergeable buffers. It reduces skb
> pre-allocations
> > and skb_frees.
> E.g. the above should go into commit message for the virtio net
> part of patchset.

Nice comments, will include them.

> I think you need to base your patch on Dave's net-next,
> it's too late to put it in 2.6.32, or even 2.6.33.
> It also should probably go in through Dave's tree because virtio part
> of
> patch is very small, while most of it deals with net/virtio_net.

> > Tests have been done for small packets, big packets
> > and mergeable buffers.
> >
> > The single netperf TCP_STREAM performance improved for host to
> guest.
> > It also reduces UDP packets drop rate.
> BTW, any numbers? Also, 2.6.32 has regressed as compared to 2.6.31.
> Did you try with Sridhar Samudrala's patch from net-next applied
> that reportedly fixes this?

Ok, I will run Dave's net-next tree.


 \ /
  Last update: 2009-12-14 21:03    [W:0.151 / U:5.852 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site