Hi,

Background:

I have a set of java background workers I start as part of my webapp. I develop locally on Ubuntu 10.10 and deploy to an Ubuntu 10.04LTS server (a media temple (ve) instance).

They're both running the same JVM: Sun JVM 1.6.0_22-b04. As part of the initialization script each worker is started with explicit Xmx, Xms, and XX:MaxPermGen settings.

Yet somehow locally all 10 workers use 250MB, while on the server they use more than 2.7GB.

I don't know how to begin to track this down. I thought the Ubuntu (and thus, kernel) version might make a difference, but I tried an old 10.04 VM and it behaves as expected.

I've noticed that the machine does not seem to ever use memory for buffer or cache (according to htop), which seems a bit strange, but perhaps normal for a server?

(edited) Some info:

(server)
root@devel:/app/axir/target# uname -a
Linux devel 2.6.18-028stab069.5 #1 SMP Tue May 18 17:26:16 MSD 2010 x86_64 GNU/Linux

(local)
wiktor@beastie:~$ uname -a
Linux beastie 2.6.35-25-generic #44-Ubuntu SMP Fri Jan 21 17:40:44 UTC 2011 x86_64 GNU/Linux

(edited) Comparing PS output: (ps -eo "ppid,pid,cmd,rss,sz,vsz")

PPID   PID CMD                           RSS    SZ         VSZ
(local)
1588  1615 java -cp axir-distribution.   25484 234382    937528
1615  1631 java -cp /home/wiktor/Code/   83472 163059    652236
1615  1657 java -cp /home/wiktor/Code/   70624  89135    356540
1615  1658 java -cp /home/wiktor/Code/   37652  77625    310500
1615  1669 java -cp /home/wiktor/Code/   38096  77733    310932
1615  1675 java -cp /home/wiktor/Code/   37420  61395    245580
1615  1684 java -cp /home/wiktor/Code/   38000  77736    310944
1615  1703 java -cp /home/wiktor/Code/   39180  78060    312240
1615  1712 java -cp /home/wiktor/Code/   38488  93882    375528
1615  1719 java -cp /home/wiktor/Code/   38312  77874    311496
1615  1726 java -cp /home/wiktor/Code/   38656  77958    311832
1615  1727 java -cp /home/wiktor/Code/   78016  89429    357716

(server)
22522 23560 java -cp axir-distribution.  24860  285196  1140784
23560 23585 java -cp /app/axir/target/a 100764  161629   646516
23560 23667 java -cp /app/axir/target/a  72408   92682   370728
23560 23670 java -cp /app/axir/target/a  39948   97671   390684
23560 23674 java -cp /app/axir/target/a  40140   81586   326344
23560 23739 java -cp /app/axir/target/a  39688   81542   326168

They look very similar. In fact, the question now is why, if I add up the virtual memory usage on the server (3.2GB) does it more closely reflect 2.4GB of memory used (according to free), yet locally the virtual memory used adds up to a much more substantial 4.7GB but only actually uses ~250MB. It seems that perhaps memory isn't being shared as aggressively. (if that's even possible)

Thank you for your help,

Wiktor