Java and alpha cluster

[Deepak Jois]

I have been assigned to debug a java program on an alpha cluster running Tru64 unix. I dont know much about it, so apologies for any incomplete info. Please let me know what else I can provide.

Typing "uname -a" gives "OSF1 atlas1 V5.1 1885 alpha"

I am runnign a few java (java with native threads) programs on the cluster and encountering a very weird problem. These java programs are basically servers listening on a socket connection. There is no client accesing them, and they remain in the sleep state most of the time.

After a while (which ranges from a few days to a few hours) the CPU usage of one of these java processes shoots up to 100% suddenly. Then, subsequently the system does not allow anybody to log in saying that there are too many users are logged in. Then after that killing these java processes makes the problem go away. I have also noticed that sometimes the CPU usage comes back to normal in about 15-20 mins and everything goes back to normal and users can log in.

My question is how may a java process with 100% usage and users not being able to log on be related. Is this some sort of a known issue. There are other java processes that run on the system and they dont cause any such problem.

I am pretty much against the wall on this one. Any help will be very appreciated. Please let me know what additional info i can provide since I do not know much about the alpha, and my job depends on it :p.

Thanks

[Ralf Puchner]

using java requires some memory settings. If the java processes listen and someone scans the ports the cpu usage will grow... that is normal. Question is if you have checked if someone tried to scan ports or connects to a specific one.

So your information is not suitable and it is time to contact the originator/developer of the java processes....

[Erich Wimmer]

Jois,
I don't know about Java, but are you sure nobody is connecting to the port your java program is listening ? It may be errorneosly an application not related to this listener. Check with "netstat -a" for connections to this port. Because not beeing able to login in case of error you should start a watching script before, for example

while true
do
date
netstat -a
sleep 120
done >/usr/tmp/netstat.log