[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: Subversion trunk (r1078338) HTTP(/WC?) performance problems?

From: Justin Erenkrantz <justin.erenkrantz_at_gmail.com>
Date: Wed, 9 Mar 2011 10:28:59 -0800

On Wed, Mar 9, 2011 at 9:12 AM, Philip Martin
<philip.martin_at_wandisco.com> wrote:
> Joe Orton <jorton_at_redhat.com> writes:
>
>> On Wed, Mar 09, 2011 at 08:50:37AM -0800, Justin Erenkrantz wrote:
>>> On Wed, Mar 9, 2011 at 8:28 AM, Philip Martin
>>> <philip.martin_at_wandisco.com> wrote:
>>> >     0.000054 epoll_wait(3, {{EPOLLOUT, {u32=30583944, u64=30583944}}}, 16, 3600000000) = 1 <0.000011>
>>>
>>> As best as I can tell, that strace isn't matching your earlier
>>> description.  The last epoll_wait is taking 0.000011 seconds?  --
>>
>> Smells like Nagle to me - looks like serf is turning Nagle *on* not
>> off... ("nodelay off" == "delay on") - try this?
>>
>> Index: outgoing.c
>> ===================================================================
>> --- outgoing.c        (revision 1440)
>> +++ outgoing.c        (working copy)
>> @@ -201,7 +201,7 @@
>>
>>          /* Disable Nagle's algorithm */
>>          if ((status = apr_socket_opt_set(skt,
>> -                                         APR_TCP_NODELAY, 0)) != APR_SUCCESS)
>> +                                         APR_TCP_NODELAY, 1)) != APR_SUCCESS)
>>              return status;
>>
>>          /* Configured. Store it into the connection now. */
>
> Yes, that's it.  Serf is now comparable to neon.

This patch is now applied to trunk and 0.7.x. Thanks! -- justin
Received on 2011-03-09 19:29:29 CET

This is an archived mail posted to the Subversion Dev mailing list.