My original posting:
Our Sparc 1000 runs Solaris 2.3, Oracle 7.0.16, Sql*Net V1, and
nothing else on 6 processors and 1 GB memory. After a reboot and
start of DB instances, vmstat shows 650 MB memory free. As the
machine approaches its regular peak load with 200 Sql connections
via Sql*Net V1, free memory drops to 350 MB. However, the similar
load on the subsequent days causes the free memory list to drop
to the level of LOTSFREE, set to 20 MB, and the kernel to start
paging out. Usually on the forth day the performance is perceived
so bad that the machine has to be rebooted.
The kernel jumbo patch (patch 101318-58?) has been applied.
Many thanks to
email@example.com (Don Lenamond)
Toby Riddell <T.J.Riddell@newcastle.ac.uk>
Graeme Robertson <firstname.lastname@example.org>
leo@Venezuela.Sun.COM (Leonardo Serres Perfeti)
Robert Sakmar <email@example.com>
most of whom suspected memory leak in some applications or the
kernel. Don Lenamond suggested the lp jumbo patch 101317-11
which would fix, among other things, lpsched's memory leak.
Toby Riddell recommended to identify the processes with increasing
sizes and talk to Sun about the kernel if none found. Graeme
Robertson was seeing the similar problem running a database called
Mapper on a Sparc 20 whereas Robert Sakmar was quite happy with
his setup on a Sparccenter 2000 with heavy Sql*Net V2 connections.
Our DBAs swiched to Sql*Net V2 and all has been fine ever since.
I would like to specially thank Leonardo Serres Perfeti of Sun who
unofficially offered his help to investigate the probelm.
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:15 CDT