Solaris HotSpot Virtual Machine Error, Unexpected Signal 11 NEED HELP

Question

Hi all
We are running a thin jsp application for video streaming live event running
on a BEA WebLogic Personalization Server 3.2 on Solaris (streaming servers
are separate and do not affect the server). After putting it in production
we have experienced serious memory problems with our environment. When the
servers are running under heavy load, they eat all available memory until
they finally crash, please see log message below. We can see that the
garbage collection is running but it doesn't seem to keep up with the memory
consumption. The problem is described in a bug report at SUN, but it is
marked as a "no fix" with the reasoning that programming resources are now
allocated to jdk1.4?
Log message:
# HotSpot Virtual Machine Error, Unexpected Signal 11
# Please report this error at
# http://java.sun.com/cgi-bin/bugreport.cgi
# Error happened during: scavenge
# Error ID: 4F533F534F4C415249530E435050079A 01
# Problematic Thread: prio=5 tid=0xdbea8 nid=0x4 runnable
When this occured we ran JDK 1.3.0 with HotSpot Server VM on Solaris 8. WLS
5.1, SP 9 and WLPS 3.2.
WLS is started with the following parameters.
#----------- Start WebLogic with the above parameters -----------
$JDK_HOME/bin/java -server -ms512m -mx512m -Duser.timezone=ECT -classpath
$JAVA_CLASSPATH -Dweblogic.system.bindAddr=10.122.114.4 -Dweblogic.cluster.m
ulticastAddress=224.0.0.100 -Dweblogic.cluster.enable=true -Dweblogic.cluste
r.name=iccluster -Dweblogic.class.path=$WEBLOGIC_CLASSPATH -Dweblogic.system
.name=$SYSTEM_NAME
weblogic.system.home=$SYSTEM_HOME -Dweblogic.home=$WEBLOGIC_HOME -Djava.secu
rity.manager -Djava.security.policy=$WEBLOGIC_HOME/weblogic.policy -Dcommerc
e.properties=$WL_COMMERCE_HOME/weblogiccommerce.properties -Dweblogic.proper
ties=$WL_COMMERCE_HOME/weblogic.properties -Dpipeline.properties=$WL_COMMERC
E_HOME/pipeline.properties -Dwebflow.properties=$WL_COMMERCE_HOME/webflow.pr
operties -Dwlcs.home=$WL_COMMERCE_HOME weblogic.Server &
We have now upgraded to jdk1.3.1_02 B02 but still have a problem with memory
leaks. We have so far been able to restart the server before memory is
exhausted and has hence not seen the same error message after the upgrade.
The memory is not freed when Weblogic aborts, the SUN server has to be
rebooted to free the memory. We now have to restart our 3 servers aprox.
every 6 hours. The servers are equiped with 2G physical memory each. We have
also tried setting hotspot to -client and setting -XX:MaxPerm.
The HW configuration is as follows:
1 Foundry Server Iron load balancer (with one hot stand by).
3 Sun E420R, 2 x 440MHz CPU, 2GB RAM. Running in a cluster with one apache
and one WLPS instance on each
The SW configuration is as follows:
JDK 1.3.1_02 B02 with HotSpot Server VM
Solaris 8.
WLS 5.1, SP 9
WLPS 3.2.
Oracle server 8.1.7
Oracle client 8.1.6?
This bug is described in
http://developer.java.sun.com/developer/bugParade/bugs/4408730.html.
It is also described as a known problem in
http://www.weblogic.com/platforms/sun/index.html.
It has also been described by SUN as a problem arising from multiple nested
try/catch blocks appearing in generated java code from servlets, especially
on WebLogic and Apache.
As SUN has said that they will not fix this bug we are trying to find a
workaround, but have yet found one. We have been looking into changing the
heap size and setting GC options on the JVM, but haven't found the optimal
solution. We would be very interested in experiences with tuning gc
parameters, documentation on solaris patches for JVMs, JVM tuning, WebLogic
tuning, HotSpot tuning.
Has anyone experienced similar problems and found a workaround. We would be
very happy to hear any suggestions as soon as possible,
thanks
Kris

Answer 1

Perhaps the -noTryBlocks option to jspc might help you. See here:
http://e-docs.bea.com/wls/docs61/////jsp/reference.html
simon.