JVM1.4.2 What can cause many full gc to occur in a row ?

Question

Hi, I'm seeing many fullgc occuring in a row on a web application runing under tomcat 5.0.28/mod_jk
System is Solaris 9, 4 proc
Java version is : 1.4.2_10
JVM Parameters are :
-server -d64
-Xmn640m -Xms2048m -Xmx2048m -XX:ParallelGC
Threads=2 -XX:+UseParallelGC -XX:+UseMPSS -noclassgc -XX:+DisableExplicitGC
-XX:MaxPermSize=128m -XX:PermSize=128m -XX:NewSize=640m
-XX:MaxNewSize=640m
(PermSize is probably too big here)
Looking at the gclog I see :
152.026: [Full GC 1334566K->986365K(1418816K), 15.1747537 secs]
169.182: [GC 1251709K->1078683K(1421056K), 1.0553477 secs]
171.721: [GC 1343868K->1125288K(1421056K), 1.2902410 secs]
173.012: [Full GC 1125288K->238719K(1418880K), 4.1125318 secs]
When adding the -XX:+PrintHeapAtGC
I see the following (cleaned-up a bit):
{Heap before GC invocations=39:
Heap
PSYoungGen      total 497728K, used 438834K
PSOldGen        total 1441792K, used 1359384K
PSPermGen       total 131072K, used 18975K
92.358: [GC 1798219K->1531712K(1940224K), 1.4728122 secs]
Heap after GC invocations=39:
Heap
PSYoungGen total 498432K, used 127090K
PSOldGen total 1441792K, used 1404621K
PSPermGen total 131072K, used 18975K
} {Heap before GC invocations=40:
Heap
PSYoungGen      total 498432K, used 467186K
PSOldGen        total 1441792K, used 1404621K
PSPermGen       total 131072K, used 18976K
95.407: [Full GC 1871808K->1448415K(1940224K), 17.1598987 secs]
Heap after GC invocations=40:
Heap
PSYoungGen total 498432K, used 6623K
PSOldGen total 1441792K, used 1441791K
PSPermGen total 131072K, used 18976K
{Heap before GC invocations=41:
Heap
PSYoungGen      total 498432K, used 340096K
PSOldGen        total 1441792K, used 1441791K
PSPermGen       total 131072K, used 18976K
113.955: [Full GC 1781887K->253003K(1940224K), 3.6430027 secs]
Heap after GC invocations=41:
Heap
PSYoungGen total 498432K, used 0K
PSOldGen total 1441792K, used 253003K
PSPermGen total 131072K, used 18976K
Now what I'm wondering is why there are 2 Full GC in a row... probably something I don't understand well in the memory collection process....
Note that this behavior happens under a load of roughly 50 concurrent users on the server
All help appreciated !
Thanks

Answer 1

Hi
Thanks for your reply,
You are right about finalizable objects. We had to add a finalize() method to an object that wasn't deallocated properly under certain circumstances.
The machine itself has about 16Gb of physical memory, but indeed at the time, we may have been in swap (each of the tomcats had about 4096 of Xmx)
Setting lower memory params (ensuring we weren't in swap) gave us better performance.
What was odd too is that we got less longer full gc when we have -Xms & and -Xmx set at different values....
We now get also in the GC log a different line, which I'm not sure how to interpret :
1488.127: [GC-- 1533225K->1533225K(1887296K), 17.6195263 secs]
Would you know what it means ? Is it another type of GC besides Full & Normal ?
I will look into the finalize as well to see if we can remove them
I will look into 5.0 as well, bu t I'm not sure we'll be allowed to put it in the production server....