Connection refused? Need help.

[stealth00]

Hi group, I am having the problem when a lot client requests come in at the same time, say 50 clients, only the first 20 or so get repsonse and all others got 'connection refused' exception. I am not sure where to specify the waiting queue time out. In my case, it's much better to let the client wait a little longer but finally get a response than let them hava a 'connection response' error. Is there a configuration file that I can modify the max time each incoming request should wait? Thank you very much!

[adrian.brock]

You are using windows?

It could be one of these:

1) You are stress testing and making lots of connections
from the same ip address. Windows refuses after about
20 attempts. It thinks you are doing a denial of
service attack.

2) From different ip addresses you don't have the
license for 50 connections. e.g. You are using
Windows XP home edition.

Regards,
Adrian

[stealth00]

Hi warjort thanks for the reply. Yeah I am doing stress test, and all the requests were from a single IP. So is there any solution to this problem? I did a search in the archive and it seems Linux doesn't have the problem, but for some reason I have to use Windows.
I also did the same thing to Weblogic6.0 on Windows, and with up to 1000 concurrent incoming requests there wasn't any problem. This did get me confused. :(

[adrian.brock]

Interesting.

I looked into registry settings on windows. I didn't find
any. :-(
I did see the same problem reported on the internet
with lots of other java applications on Windows.

I also saw the problem with the default settings of
inetd on linux, but JBoss doesn't use inetd.

Since I've got both a linux box, I just stress test on linux.

One workaround I did find, was to put a sleep
between the initial connections on the client.
I managed to get 98 out of 100 with a 1 sec sleep.

Regards,
Adrian

[stealth00]

Hi Adrian thank you for the advices, since I am doing stress testing I can't put sleep between requests :( I just searched the old posts and found someone suggested a tool called 'dynamic backlog' which is said to overcome the problem, I am wondering if there's anyone with any luck on it. Thanks again.
By the way besides the Weblogic6.0, I also tried Borland AppServer4.5 which didn't either have such a problem. I am wondering why Jboss is doing that.

[adrian.brock]

I wonder if there is some setting in rmi we are missing?

I assume you still have the other appservers installed, so
the windows registry is the same?

Regards,
Adrian

[stealth00]

Yeah I am having all three appservers installed at the same time and the other two worked just fine. Could it be the way JBoss handles clients causes this problem?

[adrian.brock]

I'll investigate some more.

I gave up last time because linux was ok.
I assumed it was a windows ip stack thing.

Of course it might be the other appservers are *slow*
at making connections, so windows doesn't think it is a
DOS attack.

Regards,
Adrian

[stealth00]

I just tried the dynamic backlog registry modification which says it will exceed the limit of windows, and no it didn't work :(

[mangomango]

Did you resolve your problem ? I have no problem with Weblogic for N>20 threads but it doesn't work on JBoss 2.4.4.
Thanks

[adrian.brock]

Hi,

I have located the cause of the problem.
Sun already know about this problem, they have it
as an open bug in Java. It only affects Windows, because
of the way it processes threads in socket handling.

There is something in JBoss I can do to work around
this problem. However the fix means the thread accepting
the RMI connection isn't the thread processing the
request. I know of at least one person using
UnicastRemoteObject to get the client's IP address.
This code would break with my fix.

I'm trying to find a different solution, but nothing
else I try works :-(

Regards,
Adrian