Re: Dead Clients with SQL*Net and TCP/IP

It happens that , at the other of the World, we suffer a similar problem awhile ago. And luckily enough, we have a solution --- but that was for Unix machines that running HPUX, SunOS, or Solaris.

Well you've have a RS6000 and I don't know whether it'll work for you. But I think the concept goes the same.

Don't believe the SQL*Net parameters can help you -- the "timeout" parameter doesn't really mean what it named. Instead, we tackle the TCP layer by cleaning sockets on the server. Unfortunately, sockets that correspond to dead Oracle shadow processes will stay around for ages until the processes are killed manually. But as you say, we don't want human intervention.

So we boldly change the TCP kernel parameters to:

• shorten the time that TCP start to look for dead connections
• reduce the number of times on retrying the connections
• shorten the idle period between each retrial.

It works well. Dead sockets are cleaned up in less than 2 minutes, comparing weeks before. Any side effect ? Yes possibly but we do not encounter one yet.

This method imply reconfiguration of the kernel. A reload of server may be required. (Solaris is an exception of this criteria.)

P.S. For anyone who interest in the solutions, please send me email or to this news group.

Andy
:) Received on Fri Sep 01 1995 - 00:00:00 CEST