Any progress on horrible IPoIB perf with kernel 4.5+?

Any progress on horrible IPoIB perf with kernel 4.5+?

[Roland Dreier]

Hi,

https://bugzilla.kernel.org/show_bug.cgi?id=111921 describes a
performance regression introduced by commit 9207f9d45b0a ("net:
preserve IP control block during GSO segmentation").  I've recently
run into this when updating my system, since that commit hit stable
kernel 4.4.y — in my case I noticed because scp over an IPoIB link
went from ~200 MB/sec down to below 1 MB/sec.  I confirmed that
reverting that change did fix my performance.

Obviously just reverting that commit isn't the right answer, since I
believe the patch description that says the commit fixes a crash.
However before I start working on this myself, I want to make sure I'm
not duplicating anyone else's work.  Doug, you mentioned having time
to work on it back in May — did you figure anything out?  Or is anyone
else investigating this?

Thanks,
  Roland
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Any progress on horrible IPoIB perf with kernel 4.5+?

[Yuval Shaia]

On Wed, Jul 06, 2016 at 10:09:51AM -0700, Roland Dreier wrote:
> Hi,
> 
> https://bugzilla.kernel.org/show_bug.cgi?id=111921 describes a
> performance regression introduced by commit 9207f9d45b0a ("net:
> preserve IP control block during GSO segmentation").  I've recently
> run into this when updating my system, since that commit hit stable
> kernel 4.4.y — in my case I noticed because scp over an IPoIB link
> went from ~200 MB/sec down to below 1 MB/sec.  I confirmed that
> reverting that change did fix my performance.

Same for us.

> 
> Obviously just reverting that commit isn't the right answer, since I
> believe the patch description that says the commit fixes a crash.
> However before I start working on this myself, I want to make sure I'm
> not duplicating anyone else's work.  Doug, you mentioned having time
> to work on it back in May — did you figure anything out?  Or is anyone
> else investigating this?

Adding Haakon which as far as i recall has some insights.

> 
> Thanks,
>   Roland
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Any progress on horrible IPoIB perf with kernel 4.5+?

[Doug Ledford]

[-- Attachment #1.1: Type: text/plain, Size: 1402 bytes --]

On 7/6/2016 1:09 PM, Roland Dreier wrote:
> Hi,
> 
> https://bugzilla.kernel.org/show_bug.cgi?id=111921 describes a
> performance regression introduced by commit 9207f9d45b0a ("net:
> preserve IP control block during GSO segmentation").  I've recently
> run into this when updating my system, since that commit hit stable
> kernel 4.4.y — in my case I noticed because scp over an IPoIB link
> went from ~200 MB/sec down to below 1 MB/sec.  I confirmed that
> reverting that change did fix my performance.
> 
> Obviously just reverting that commit isn't the right answer, since I
> believe the patch description that says the commit fixes a crash.
> However before I start working on this myself, I want to make sure I'm
> not duplicating anyone else's work.  Doug, you mentioned having time
> to work on it back in May — did you figure anything out?  Or is anyone
> else investigating this?
> 
> Thanks,
>   Roland
> 

I *just* came back from PTO.  I'm still working through my email (where
I saw your writeup of this issue).  I didn't get very far with the issue
back in May, but that's because I didn't have a lot of the background
information you supplied in the other thread and didn't have the time to
dig it out on my own.  I could probably make some progress now.

-- 
Doug Ledford <dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>
    GPG Key ID: 0E572FDD


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]