[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: Justin Erenkrantz <justin_at_erenkrantz.com>
Date: Thu, 12 May 2011 11:37:02 -0700

On Thu, May 12, 2011 at 11:02 AM, Mark Phippard <markphip_at_gmail.com> wrote:
> On Thu, May 12, 2011 at 1:57 PM, Justin Erenkrantz
> <justin_at_erenkrantz.com> wrote:
>>> Wasn't the patch for mod_deflate?  And just to fix the memory leak
>>> when a client that does not support gzip connects?
>>
>> Are you by chance using SSL?
>
> No.

Hmm. More specifics on your configuration would be helpful if I am to
try to reproduce your setup. Serf version? (0.7.2 or higher would be
needed to avoid the Nagle issue.) OS (client & server)? You
mentioned on the page that you are using a VPN - what type of VPN?
It's highly likely some VPNs might play havoc with the OS networking
stack.

Since I can reproduce, I'll look into the OpenSSL stuff - since
ra_neon doesn't seem to have a huge hit, I'm hoping it is pretty
straightforward.

> I was mainly trying to see the HTTP request count with HTTPv2 so I a
> using a 1.7 server.

I maintain that request count isn't an highly interesting metric -
especially given serf's design. Since the requests are pipelined, the
latency impacts should be minimal - if it were serialized and
synchronous, yes, it'd be a concern. But, it's parallel and async, so
that number doesn't dominate in a properly tuned environment - what
you give up by sending more from the client, you should get back from
having better server-side scalability/partitioning capabilities. --
justin
Received on 2011-05-12 20:37:29 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.