Negative timestamps in GC verbose log

Question

Hi,
I am experiencing a strange problem with my GC logs - some of the GC timings are negative:
2008-07-31T12:01:38.223+0100: 1289.832: [GC 1289.832: [ParNew: 240320K->21823K(240320K), -0.2231308 secs] 703715K->519899K(1288896K), -0.2229563 secs] [Times: user=0.47 sys=0.00, real=0.28 secs]
857273K->684381K(1288896K), 0.1177372 secs] [Times: user=1.17 sys=0.00, real=0.24 secs]
2008-07-31T12:01:58.769+0100: 1309.863: [GC 1309.863: [ParNew: 240320K->21824K(240320K), 0.5649079 secs] 902877K->729310K(1288896K), 0.5650772 secs] [Times: user=0.45 sys=0.00, real=0.25 secs]
2008-07-31T12:02:00.754+0100: 1312.161: [GC 1312.161: [ParNew: 240320K->21824K(240320K), -0.2063372 secs] 947806K->746855K(1288896K), -0.2061724 secs] [Times: user=0.30 sys=0.02, real=0.09 secs]
2008-07-31T12:02:03.832+0100: 1315.251: [GC 1315.251: [ParNew: 240058K->21823K(240320K), -0.1469380 secs] 965089K->771703K(1288896K), -0.1467691 secs] [Times: user=0.39 sys=0.00, real=0.17 secs]
I am running java 6 update 4 on a windows machine with 8 AMD cores, 16Gb of RAM, Windows Server 2003 R2 SP1.
The problem can only be solved by restarting the Windows server - restarting the JVM doesn't do anything for this.
I am worried that these negative timings might affect the internal garbage collection algorithms although I don't have any way to know that for sure.
My GC tuning parameters are:
param00= -server
param01= -XX:+UseConcMarkSweepGC
param02= -XX:SurvivorRatio=10
param03= -XX:NewSize=256m
param04= -XX:MaxNewSize=256m
param05= -XX:CMSInitiatingOccupancyFraction=76
param06= -XX:ParallelGCThreads=8
param07= -XX:+UseParNewGC
param08= -XX:-DisableExplicitGC
param09= -XX:+UseCMSCompactAtFullCollection
param10= -Xms1280m
param11= -Xmx1280m
param12= -XX:+PrintGCDetails
param13= -XX:+PrintGCDateStamps
Has anyone seen this behaviour before?
Can anyone at Sun confirm whether the negative timestamps affect the internal GC logic?
Many thanks,
Michele

Answer 1

mwertejuk wrote:
Have look at this bugreport:
http://forums.java.net/jive/thread.jspa?threadID=61998
It's probably the same issue and not fixed in 137138-09 but in 139556-08.
I've noticed that this error disappears if I use different GC logging parameters, but I assume it's safer to do the solaris upgrade to 139556-08.As noted it is a Solaris, not VM, problem.