Jon Stevens wrote:

> Ok, I put a whole bunch of logging into Turbine to see *exactly* what line
> of code is causing the slowness that I keep reporting here and I have now
> found it...
>
>         Log.note ("RunDataFactory: 11");
>         // Get the HttpSession object.
>         data.setSession ( data.getRequest().getSession(true) );
>         Log.note ("RunDataFactory: 12");
>
> As you can see above, essentially, all that is happening is that I'm storing
> an instance of the HttpSession object within the RunData object. Marking
> things as "true" causes the redirect to happen, so there is another
> request...
>
> [Thu Dec 21 13:34:15 PST 2000] -- NOTICE  -- RunDataFactory: 11
> [Thu Dec 21 13:35:01 PST 2000] -- NOTICE  -- RunDataFactory: 12
>

This is the first session create since the webapp was started, right?
Currently, that is when the RNG is initialized.

> ...
> [Thu Dec 21 13:35:03 PST 2000] -- NOTICE  -- RunDataFactory: 11
> [Thu Dec 21 13:35:03 PST 2000] -- NOTICE  -- RunDataFactory: 12
>

And this is the behavior you would see after the first one.

>
> As you can see above the first request through this code takes bloody
> FOREVER and the second one is quite fast.
>
> The really *INSANE* part about all of this is that people have checked
> Scarab out of CVS themselves on the SAME JVM (1.3 on Windows) and don't see
> any real slowness at all (approx 4-5 seconds).
>

That's definitely strange ... 4-5 seconds is my experience even on much slower
hardware, on both 1.2.2 and 1.3.

>
> So, Craig, can we please try to do something about this? There has got to be
> something wrong with either my setup or something else (I really don't think
> this is entirely a classloader issue anymore). I also have this great
> slowdown on MacOSX as well.
>

This part isn't.  Aleady on my TODO list is moving the RNG initialization to
webapp startup time -- that will normally hide it for a 4-5 second delay, but
certainly won't fix a 45-second delay.

>
> If you want to duplicate it, you can check Scarab out of CVS and do it
> yourself. I have re-done the CVS tree and it is very easy to get things up
> and running (even without a database installed...just ignore that part of
> the README.txt file).
>
> <http://scarab.tigris.org/>
>

You mean it's not just "check out, compile, and run"???  :-) :-)

Will do.


>
> thanks,
>
> -jon

Craig


Reply via email to