Re: Serf takes more time to complete the tests than dav

On 12/29/06, Justin Erenkrantz <justin@erenkrantz.com> wrote:
> I haven't had time to dig down and figure out why yet. I may get to
> it this weekend, time permitting.

Actually, this might have been really trivial.

Kamesh: Please try out r1097 of serf. This has a fix for being
smarter about writing headers. This greatly reduces the amount of TCP
packets serf puts out. If this helps, I'll cut a serf 0.1.1 with this
fix included.

Just for basic tests #1, it cut the captured network traffic from
~500KB down to ~140KB - bringing the number of packets almost
identical to ra_dav. (I think loopback hosts are more sensitive to
this sort of thing, but reducing the overall number of packets is
generally a good thing!)

Here's what I have now:

basic tests with ra_dav:
real 2m15.188s
user 0m21.034s
sys 0m18.950s
(2.8MB of traffic)

basic tests with ra_serf (r1097 of serf):
real 2m14.056s
user 0m18.697s
sys 0m19.278s
(3.3MB of traffic)

basic tests with ra_serf (pre-r1097)
real 2m24.351s
user 0m18.817s
sys 0m19.923s
(12MB of traffic)

Thanks. -- justin

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Dec 29 18:46:59 2006