On Thu, Jun 28, 2001 at 11:21:52PM -0400, Donald Ball wrote:
> i care very much about mod_webapp. er, well, to be more clear, i care very
> much about having an apache-tomcat connector module that doesn't require
> me to put the same configuration data in two completely different places.
> i presume mod_webapp is the only component trying to tackle that job, but
> if there are others, i'm all ears.
As I see it, that is the big user difference. But, I'm not familiar
enough with the underlying protocols (warp v. ajp) to tell you from the
network level how different they are. I could be wrong here.
> anyway, i just ran some benchmarks and mod_webapp hasn't choked once, but
> it is significantly slower than tomcat standalone. for 100 requests at
> concurrency 10, it typically takes apache+tomcat 45 seconds, while tomcat
> gets through at about 30 seconds. logging could account for some of this,
> as mod_webapp and/or the java warp connector seem to be very verbose.
> nonetheless, do others experience this?
Tomcat's HTTP standalone connector isn't very good at parsing HTTP
(but, I know the HTTP protocol too well to know where the pitfalls are).
IMHO, Tomcat shouldn't be allowed to be a web server on its own. It
needs a connector in front of it. So, just from a standards compliance
standpoint, Tomcat's HTTP connector has major issues with it. ASF has a
very fine Web server available, so Tomcat doesn't need to be a HTTP
server.
And, yes, mod_webapp is WAY too verbose. I hope that will be
addressed soon (or I'll submit a patch to do so when I can get around
to it). The other advantage of using mod_webapp (or mod_jk) is that
you can now serve static pages from Apache which should blow away
Tomcat w.r.t. speed. My bet is that Pier has just focused on
functionality, not tuning the Warp code for performance.
> oh, also, i ran into an OutOfMemory error when i really slammed the
> server. i hadn't experienced that before with tomcat standalone. could
> there be a memory leak in the WarpConnector code? has anyone else run into
> this?
Hmm, I haven't seen anything, but I also haven't done massive load tests
on it. I thought Java wasn't supposed to have memory leaks? -- justin
