Hi Vinod,
first of all many thanks for your respond :)
Now once more to my question / problem:
The 4 seconds you have mentioned can I measure too, but(!) this time
doesn't include the ConnectFuture.join() method, right? When I wait for
each connection until it joined correctly, I measure 450 seconds in
worst case instead. I've tried some different settings (more threads
than one) in the SocketAcceptor too and could reduce the time to 350
seconds.
But with the old implementation 0.9.5 this time was definitely better
(120 seconds).
I should mention, that I use the original Java5 concurrent stuff instead
of the backport lib.
Should I make an JIRA entry? Don't know.
Best Regards
Michael
Vinod Panicker schrieb:
Hi Michael,
On 10/19/06, Michael Bauroth <[EMAIL PROTECTED]> wrote:
Hi,
I was very busy for the last few months so that I had not the chance to
read much about the Mina progress in this time.
Today I've checked out the latest version and compared it's performance
against 0.9.5. It seems that it consumes less cpu power but more memory.
Is this correct so?
Surprising you say that, since in my recent load tests, I was seeing
huge memory consumption which I had attributed to my code. Will have
to take a second look now.
Two questions about the changes: Is now per default the Executor
Threadpool enabled in the acceptor? And the second question: Could there
be a reason, that establishing of 20000 concurrent connections now needs
400 seconds against 100 seconds before?
AFAIK, yes to Q 1. WRT Q 2, I've been establishing 20k concurrent
connections too, but its @ 4/sec, so dunno about the difference.
I think we need to do some memory profiling ASAP.
Regards,
Vinod.
