Emmanuel,
Awesome! That seems to have helped quite a bit.
Have you thought about using a profiler during the load test? I think MINA
has a YourKit Open Source license. Are we allowed to use that? It would be
good to find out exactly where the hot spots are and figure out whether
something in MINA could be optimized.
I'd do it myself, but I'm more focused on the datagram transport right now.
I'd like to use YourKit to profile it as well, as I hope to contribute it to
MINA before 2.0.0.
Also, what JVM are you using?
-Greg
On 3/30/07, Emmanuel Lecharny <[EMAIL PROTECTED]> wrote:
Oops, I just forgot the reply all...
---------- Forwarded message ----------
From: Emmanuel Lecharny <[EMAIL PROTECTED]>
To: Greg Duffy <[EMAIL PROTECTED]>
Date: Sat, 31 Mar 2007 00:34:40 +0200
Subject: Re: High latency in Mina ?
Greg,
I have done some more test, on the 'big server' ( 8 CPUs), changing the
SocketAcceptor initialization with something like what you suggested :
tcpAcceptor = new SocketAcceptor(
Runtime.getRuntime().availableProcessors(),
threadPoolExecutor );
(we already have this threadPoolExecutor, so I reuse it).
The result is huge : I can get as much as 6400 request/s, instead of
only 2000 previously ! And the CPU reach 85% of user+syst usage (70%
user, 10 to 15% system)
This is just what I expected :)
I have more tests to do, but this is a big step in the good direction !
Thanks a lot !
Emmanuel
