Question

I am performing tuning benchmarks on a web application that is running on JBoss AS5.

I am creating different scenarios using JMeter, starting from low load till stress load.

What I have noticed is that the GC log starts always with a full GC.

Can someone explain to me this behavior?

Thanks in advance.

FWIW, here is a snippet from a GC log:

17.560: [Full GC [PSYoungGen: 44456K->0K(458752K)] [ParOldGen: 0K->4385K(1572864K)]      44456K->4385K(2031616K) [PSPermGen: 11565K->11555K(262144K)], 0.9226691 secs]
72.478: [GC [PSYoungGen: 393216K->30720K(458752K)] 397601K->35105K(2031616K), 0.1787110    secs]
112.137: [GC [PSYoungGen: 423936K->38912K(458752K)] 428321K->43297K(2031616K), 0.2197971 secs]
188.297: [GC [PSYoungGen: 432128K->54272K(458752K)] 436513K->58657K(2031616K), 0.3034273 secs]
309.100: [GC [PSYoungGen: 447488K->60416K(458752K)] 451873K->64801K(2031616K), 0.3111470 secs]
430.354: [GC [PSYoungGen: 453632K->65536K(454848K)] 458017K->72129K(2027712K), 0.3374716 secs]
546.078: [GC [PSYoungGen: 454848K->65536K(415104K)] 461441K->78881K(1987968K), 0.3746511 secs]
652.116: [GC [PSYoungGen: 415104K->40960K(436928K)] 428449K->88641K(2009792K), 0.3895185 secs]
765.134: [GC [PSYoungGen: 390528K->28672K(437632K)] 438209K->94882K(2010496K), 0.2703870 secs]
870.726: [GC [PSYoungGen: 380800K->23552K(375680K)] 447010K->102114K(1948544K), 0.1948568 secs]
976.144: [GC [PSYoungGen: 375680K->18432K(436096K)] 454242K->110306K(2008960K), 0.1734677 secs]

Answer 1

With the help of Peter Johnson from the Jboss Community. I’ve been able to know the answer of this question.

I am writing it here to share it with you.

The call to System.gc() is coded into the app server (or into one of the libraries it uses – I know the the RMI libraries like to make GC calls), I suspect as it transitions from bootstrap to the app server itself. I think it might also perform another GC or two after deployment or other major initialization activity. I think the idea behind this was to have the heap as clean of junk as possible before handling user requests. Of course, that made sense years ago when when servers didn’t have that much memory and heaps were smaller. I have seen this behavior in JBoss AS release from 4.0.x up through 5.x (haven’t monitored GC behavior on 6).