Hi,

I am just about to place into CVS what I think is
very close to a 0.5 milestone release of Velocity.
But I could really use some eyeballs to find one
particularly nasty problem.

I am using the same settings that Peng Yong has been
using with ApacheBench

./ab -n 1000 -c 20 url

During the first round of testing is where the problem
is. I am using a single template but the template loader
is creating two instances of the template and I believe
the cache is getting confused. Anyway the performance
during the first phase of testing is not good due to
this internal confusion. But if I give Velocity a
short rest every subsequent round of testing is fantastic.
But everytime I restart I get the same initial problem.

What I am consistently seeing is Velocity being 20-30%
faster then WM _everytime_ after the first round. And
the load monitor shows that Velocity uses far less of
the system resources. If someone can help me locate the
initial startup problem we will have something
substantially faster then WM. Once this is fixed then
I can run some long-lived benchmarks and publish the
results.

There is also a couple of parsing problems that need
to be fixed. Geir and I are looking at them and they
should be resolved by the end of day.

So if this startup problem is identified and the
last couple parsing problems are fixed. I would
be willing to release the version of new torque
I have here working with Velocity and announce
the Velocity Service for Turbine. Along with a
few other cleanups after that I would be willing
to call it a 0.5 milestone release.

It has been roughly 10 weeks and I think we
have something performs better then WM, and
with a little work the code can be better:
it is definitely smaller code base and as a
result much easier to digest. Geir was able
to dig in very quickly! :-)

Anyway I will start placing the code in CVS and
if someone can help me identify the startup
problem I would greatly appreciate it!

I believe the problem is occuring in the template
loader and the template caching system.

jvz.

---

Jason van Zyl
[EMAIL PROTECTED]

Reply via email to