[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CA+5PVA4YmFVy=02Py2+1=BP4MbcCrUKphk7w+0-ahvMPxygt_g@mail.gmail.com>
Date: Wed, 11 Sep 2013 21:51:58 -0400
From: Josh Boyer <jwboyer@...oraproject.org>
To: Yuchung Cheng <ycheng@...gle.com>
Cc: Michael Sterrett <michael@...rretts.net>,
netdev <netdev@...r.kernel.org>,
Neal Cardwell <ncardwell@...gle.com>,
Nandita Dukkipati <nanditad@...gle.com>
Subject: Re: 3.10.0 network trace
On Wed, Sep 11, 2013 at 12:32 PM, Yuchung Cheng <ycheng@...gle.com> wrote:
> On Wed, Sep 11, 2013 at 5:49 AM, Josh Boyer <jwboyer@...oraproject.org> wrote:
>> On Sun, Jul 21, 2013 at 9:36 PM, Michael Sterrett <michael@...rretts.net> wrote:
>>> Since upgraded to 3.10.1 which exhibits the same issue (probably not
>>> surprising).
>>>
>>> sysctl net.ipv4.tcp_frto=0 net.ipv4.tcp_early_retrans=3 seems to take
>>> care of it.
>>
>> We're still getting reports of this with 3.10.10 in Fedora [1]. At
>> least one of the reporters has said those sysctl settings didn't help.
>>
>> Is there anything else that can be gathered to help track this down?
> since frto=0 didn't help, I suspect it's either a bug introduced in
> the tail loss probe patch 9b717a8d (tcp: TLP loss detection) or other
> changes related loss recovery.
>
> could you try disable TLP by
> sysctl net.ipv4.tcp_frto=0 net.ipv4.tcp_early_retrans=1
>
> if it still does not work. disable any form of early retransmit by
> sysctl net.ipv4.tcp_frto=0 net.ipv4.tcp_early_retrans=0
Thanks. I've asked the people in the bug to try those settings.
> if you can test a custom build kernel, please try this debugging patch.
I'll get the patch included in a test build for the tomorrow as well.
josh
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists