Closed as program error.
Error Message: Verbosegc Output showing huge errant values for the attributes tracetarget and kickoff. Here's a sample: <con event="kickoff" timestamp="Sat Oct 31 11:34:45 2009"> <stats tenurefreebytes="0" nurseryfreebytes="112131288" tracetarget="10145709460442578944" kickoff="1268213682555469598" tracerate="1961.12" /> </con> Also an outage for about 5 minutes is observed usually when these entries are seen in the verbosegc output. . Stack Trace: N/A .
There is no real work around. The only solution is to run with -Xgcpolicy:opthtruput or to get a VM with the fix.
This problem is specific to GC policies using concurrent collector (that is, gencon or optavgpause). The errant verbosegc entries were caused by a divide by 0. A divide by 0 when using floating point values results in a NAN. When this value is converted back to an unsigned integer the number is very large. This large number would cause the GC to kick off another concurrent collection immediately since the amount of free memory would never be larger than this errant number, resulting in a consecutive concurrent collections for a few times and hence an outage in the customer environment.
This defect will be fixed in: 5.0.0 SR12 . Concurrent GC code has been updated to check to see how much trace work was completed concurrently and if this value is 0 then do not adjust the kick off value. . To obtain the fix: Install build 20100204 or later
Reported component name
JAVA 5 J9 COMMO
Reported component ID
Last modified date
APAR is sysrouted FROM one or more of the following:
APAR is sysrouted TO one or more of the following:
Fixed component name
JAVA 5 J9 COMMO
Fixed component ID
Applicable component levels