JDBC thin driver TCP TIME_WAIT on ORACLE 8.1.7.2.1 (win2k server)

Hi,

We are running IBM's WebSphere Application Server on several servers. They all use the same Oracle 8.1.7.2.1 database; the WebSphere servers have a Oracle 8.1.7.2.1 client installed but use the JDBC thin driver to connect to the database.

We are running in a production environment; since using WebSphere, we see an enormous amount of TCP TIME_WAIT connections using netstat on the Oracle server. Most of them (around 800) are of the following form (wolnlztmpor01 is a WebSphere server, wolnlztmora01 is our database server and 1521 is its listener port):

TCP wolnlztmora01:1521 WOLNLZTMPOR01:1741 TIME_WAIT

We also see around 80 TCP connections like this (x and y are both ports>1024):

TCP 127.0.0.1:x 127.0.0.1:y TIME_WAIT

Closing all connections from the WebSphere server by stopping the java processes (and hence all JDBC connections) does not change anything.

I have been studying some material on this topic, from which I conclude that on win2k server these TIME_WAIT connections should cease to exist after 240 seconds (configurable through HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\Parameters\TcpTimedWaitDelay). Does it make sense to use a lower value here (60 seconds for example)?

I also see people who recommend to stop the ORACLE listener process on a regular basis and clean up some ORACLE logs on the filesystem.

Perhaps this is a known problem... is there a patch available?

Please help, we are facing a deadline here.

Regards,
Hendrik Received on Fri Jan 23 2004 - 05:16:04 CST