Jackrabbit Lucene indexes

Question

I have Jackrabbit 2.4.0 (deployed as rar into a JBoss AS 7.1.0) on a Red Hat 6 64-bit machine. The JBoss JVM has the max heap size set to 8 GB. The machine has 24GB of RAM. The curious thing is when the JBoss is started it has almost 20 GB of virtual size (statistic taken from top). The Linux page cache (swap cache) is around 10 GB so the system will actually have free memory somewhere around 5 GB. I'm not sure why the page cache is so big and I'm trying to make a link with the size of the jackrabbit's data directory where the Lucene indexes are kept. The size of the directory is around 10 GB. My question is: does Lucene use memory mapping for indexes files?

Thanks in advance.

Answer 1

My question is: does Lucene use memory mapping for indexes files?

Lucene uses memory mapping only if you ask for it. You might also want to read this .

Answer 2

JBOSS total resident memory is comprised of several factors, not all of which are heap:

1. Heap
2. Perm gen
3. JARs and JVM
4. Mapped byte arrays
5. Thread stacks (~1MB per thread)

There's your application and the app server itself.

So certainly you should profile using something like Visual VM to see the details of what's going on, but it'll only help with heap and perm gen. You'll need tools like nmap on Linux to ferret out the rest.