IBM JVM 1.3 cx130 runs out of memory after running fine for hours.

[Michael Stack]

Having run sustained loads on cx130-20001124 and the previous version of
the IBM JVM 1.3, cx130-20000815 I have noticed that the IBM JVM runs
fine for hours under an unvarying, sustained load but then of a sudden
memory stops being freed. Even when the load is removed, the JVM is
frozen in its occupied memory state. It never finds the wherewithal to
free its heap.

This scenario is reproducible. It does not happen w/ the Hotspot JVM.

The load was being put on an SMP machine. I have not tried it against a
single processor machine. IBM are ambiguous about whether the IBM JVM
runs on SMP. There is no explicit denial that the JVM does not work on
SMP up on the IBM site. The list of fixes for the latest JVM even notes
a fix for hangs on SMP (See extract below). But then see elsewhere in
this newsgroup, the message "Re: IBM JDK 1.3 + Tomcat + SMP = Disaster"
from Tue, 09 Jan 2001 14:29:42 where an IBM representative says "SMP is
not currently supported in the 1.3 JDK, but extensive development work
has been done so to do."

Extract from IBM list of fixes for latest JVM release. From:
http://207.25.253.61/downloads/reg/devcjjd/%3Cd62696b636164737%3Espdxwt7d3annghmbme72279t/developer/java/linux130/fixes.html

+--------------+------+-------+--------+----------------------------------------------+

|Build |Defect|APAR
|SunBugId|Description |
+--------------+------+-------+--------+----------------------------------------------+

|GA Release
cx130-20000623
|
+--------------+------+-------+--------+----------------------------------------------+

|cx130-20000706|13323 | |N/A |Occasional hang on
SMP |
+--------------+------+-------+--------+----------------------------------------------+

Do others observe the above w/ the IBM JVM?

Our application is a bunch of servlets running inside JServ.

Thanks,
St.Ack

[Neil Masson]

Can you supply a testcase for ths?

Concerning SMP support, it's certainly unclear from the Web site what the
situation is. I am sure we don't claim support for SMP for Linux, though.
The situation is a little complicated as the statement that not supported
on SMP is not the same as saying doesn't run on SMP. I myself use an
SMP box, but was forced to run in single-processor mode until glibc
2.1.3-21 came along. With this library most Java tasks run well enough,
but there are a number of test cases which hang the JVM.

As I have said previously there are a number of development fixes in the
threading and signalling area which have not yet been released into the
service code base. SMP support remains a goal.

The bug fix you noticed is a red herring as it refers to some code common
to other platforms and is not relevant to the specific Linux problems.

Neil Masson
IBM Java Technology Centre