Re: SQLNET w/Multinet

From: Frank Newman <fnewman_at_bdmserver.bdm.com>
Date: 18 Aug 1994 21:15:31 GMT
Message-ID: <330j1j$nok_at_burrito.abq.bdm.com>


Jennifer R. Amon (bamon_at_ocvaxc.cc.oberlin.edu) wrote:
: Any problems? We're just moving to Multinet, and I'm reading about
: problems over on the ORACLE-L listserv.
 

: Thanks for any info.
 

: _____________________________________________________________________
: Jennifer R. Amon PHONE: (216) 775-6987
: Houck Computing Center FAX: (216) 775-8573
: Oberlin College
: Oberlin, OH 44074 INTERNET: bamon_at_ocvaxc.cc.oberlin.edu
: _____________________________________________________________________

We are running SQL*Net 1.2 with Multinet version 3.3 on a Digital Alpha machine. Generally, it works okay except if a PC client abnormally terminates an SQL*Net connection (such as a program crash), the IP socket and sometimes the server process continue to run. In this case, Oracle noramally relies on the transport vendor to kill the IP socket using the keepalive option. Unfortunately, this option is normally turned off with TGV's implementation. While TGV provides the option to turn on keepalive with their API, it doesn't seem that Oracle has taken advantage of it. As a result, motherless IP sockets continue to send ack messages back and forth until the socket is killed off at the PC end. These motherless IP connections tend to pile up after a while, slowing things down. If, in addition, the Oracle server process for that socket does not also get killed off, then you could also exceed your Oracle license limit even though only 1 or 2 users may be actively connected. Right now, we are managing this problem by asking developers to make sure they kill off any and all connections when they leave for the day and using the multinet show/connections command to find and warn users who have a lot of connections (more than 2). Good luck. Received on Thu Aug 18 1994 - 23:15:31 CEST

Original text of this message