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

Re: 1.7 Performance via HTTP

From: Mark Phippard <markphip_at_gmail.com>
Date: Thu, 12 May 2011 10:23:02 -0400

On Thu, May 12, 2011 at 2:12 AM, Justin Erenkrantz
<justin_at_erenkrantz.com> wrote:
> On Wed, May 11, 2011 at 4:40 PM, Mark Phippard <markphip_at_gmail.com> wrote:
>> The story using Serf is not as good.  There are a few places where it
>> is fastest, namely merge.  But there are other cases where it is
>> dramatically slower.  The number of HTTP requests with Serf is 80,990.
>>  Looking at the numbers, it seems like Serf is slowest when it comes
>> to the areas where it issues all those GET requests, such as checkout
>> and update.  On other operations it is more inline with Neon.  Maybe
>
> r1102173 will help slightly when there are lots of checkouts as
> ra_serf wasn't asking for the REPORT response to be compressed - doing
> that saves a fair bit of traffic before the parallelization can begin.
>  (86k down to 16k for a checkout of notes/.)
>
> I'm also assuming you've appropriately configured your httpd server...see:
>
> http://mail-archives.apache.org/mod_mbox/subversion-dev/201101.mbox/%3CAANLkTi=AN-2BZgjSioowmEFgqFt5L=wyW9yzfFayO6Xy@mail.gmail.com%3E
>
> FWIW, in some quick local real-world tests, I see ra_neon and ra_serf
> being within margin of error on checkouts - so I have a hunch it might
> be something more pathological that a test suite would hit.  I'm sure
> those of us in Dublin or Berlin can look more.  I'll be in Dublin this
> weekend, but I won't be in Berlin...  -- justin

I re-ran the results with your latest changes and posted the new numbers.

https://ctf.open.collab.net/sf/go/wiki2118

I also turned on mod_deflate. I do not normally use it for the reason
Ivan mentioned.

I also changed MaxKeepAliveRequests from 64 to 5000

The numbers are a little better. There is on anomaly in the
FolderTests where an update blew up from 2 seconds to over 10 minutes.
 I had this problem one other time when running with Serf. I am not
sure what goes on when that happens. Maybe something similar happened
with the initial checkout when it went from 20 minutes to 2 minutes?

I am currently running the Neon tests again to while I have
mod_deflate configured.

-- 
Thanks
Mark Phippard
http://markphip.blogspot.com/
Received on 2011-05-12 16:23:31 CEST

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

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.