Threaded execution (was: Interesting Bugs in 12cR1)
From: Yong Huang <yong321_at_yahoo.com>
Date: Fri, 12 Jul 2013 08:48:21 -0700 (PDT)
Message-ID: <1373644101.10961.YahooMailNeo_at_web184804.mail.gq1.yahoo.com>
> I've been looking at threaded execution and came across the sys auth issue,
> to date I've yet to find any great benefit for this process model, which
> isn't the default.
I did a quick test on the performance of establishing and disconnecting local sessions on a not very fast RHEL 5 box. With threaded_execution, making 200 sessions takes 2 seconds. Without, it takes 3 seconds. The test is done more than 20 times and the first measurement is discarded. The timing is accurate, rarely off by 1 second. I think this execution model is meant to make fork (clone) and exec more lightweight. For any reason you can't use shared server config, this is an alternative, so you still have a dedicated session and its own trace file, etc. The shadow "process" becomes shadow thread. (But I remember the number of Linux threads counts against user's process limit, i.e. ulimit -u.)
Date: Fri, 12 Jul 2013 08:48:21 -0700 (PDT)
Message-ID: <1373644101.10961.YahooMailNeo_at_web184804.mail.gq1.yahoo.com>
> I've been looking at threaded execution and came across the sys auth issue,
> to date I've yet to find any great benefit for this process model, which
> isn't the default.
I did a quick test on the performance of establishing and disconnecting local sessions on a not very fast RHEL 5 box. With threaded_execution, making 200 sessions takes 2 seconds. Without, it takes 3 seconds. The test is done more than 20 times and the first measurement is discarded. The timing is accurate, rarely off by 1 second. I think this execution model is meant to make fork (clone) and exec more lightweight. For any reason you can't use shared server config, this is an alternative, so you still have a dedicated session and its own trace file, etc. The shadow "process" becomes shadow thread. (But I remember the number of Linux threads counts against user's process limit, i.e. ulimit -u.)
Yong Huang
-- http://www.freelists.org/webpage/oracle-lReceived on Fri Jul 12 2013 - 17:48:21 CEST