Re: stress test probs

[Roger Dingledine <arma@mit.edu>]

On Wed, Aug 28, 2002 at 02:40:13PM -0400, Bruce Montrose wrote:
> I noticed that there were pages of httpap processes after the test completed.
> 
> Might this be problematic if the onion router needs to handle numerous 
> simultaneous connections?

This is the case because the httpap we're using forks for each new
connection. It will certainly be a bottleneck if we start wanting to
get lots of performance out of the system.

We should switch over to using Privoxy, which uses threads. It's on the
todo list. :)

> My testing starts off with 16 simultaneous users and steadily ramps up to 
> 256 simultaneous users.

With a bunch of simultaneous users, is it more realistic to think each
user or a few users will have an httpap on their own machine, so that's
not a good measure of how well the onion routers can handle things?

When I was playing with httperf, I ran the httperf on one machine, the
httpap's on a second, and the onion routers on different ones. Still, at
that load we may still see the httpap machine being the bottleneck. The
real answer is to switch to something like squid or privoxy, and even
then it'll be a tossup for which one will be the bottleneck (I guess a
tossup is better than knowing for sure).

--Roger