Eclipse Community Forums: Memory Analyzer (MAT)

Home » Archived » Memory Analyzer (MAT) » Using pre-indexed dump

Wed, 15 October 2008 16:24

Eclipse User

I indexed a big heap dump on a 64-bit machine with lots of memory, using
the advice from a previous posting ("Performing initial
analysis/conversion using command line tool").

The only place I can run the GUI is on a 32-bit machine, so I copied the
index files over to the 32-bit machine and ran MemoryAnalyzer. However,
when I open the heap dump file, MemoryAnalyzer insists on re-indexing the
dump, which fails because the dump is too big to be parsed on a 32-bit
machine.

How do I get MemoryAnalyzer to use the pre-created indexes? Or
alternatively, how do I get MemoryAnalyzer to create an X window so I can
run it on the 64-bit machine but see the GUI on the 32-bit machine (I can
easily open xterms from the 64-bit machine, so X is working properly, but
MemoryAnalyzer immediately exits).

Indexes created on a 64-bit rhel4 machine, GUI run on a 32-bit rhel5
machine.

Re: Using pre-indexed dump [message #5459 is a reply to message #5452]

Wed, 15 October 2008 20:21

Eclipse User

Never mind about the X part, I got that to work (I just had to use the
64-bit version of MAT on the 64-bit machine).

I am still curious about getting MAT to use the indexes of a pre-indexed
dump.

Re: Using pre-indexed dump [message #5479 is a reply to message #5452]

Fri, 17 October 2008 03:24

Eclipse User

B R wrote:

> I indexed a big heap dump on a 64-bit machine with lots of memory, using
> the advice from a previous posting ("Performing initial
> analysis/conversion using command line tool").

> The only place I can run the GUI is on a 32-bit machine, so I copied the
> index files over to the 32-bit machine and ran MemoryAnalyzer. However,
> when I open the heap dump file, MemoryAnalyzer insists on re-indexing the
> dump, which fails because the dump is too big to be parsed on a 32-bit
> machine.

> How do I get MemoryAnalyzer to use the pre-created indexes?

Normally this is the way I also work with larger dumps - parse them on a
bigger box and then copy back to my PC all the files. So this is the right
way to go. At the end of the parsing we serialize some data in one of the
files. If later we fail to deserialize it, we trigger re-indexing. This
could happend if the MAT versions on both machines differ. But I don't
remember intruducing any changes which can trigger this recently. Any
error messages?

Can you look for and send any error messages from the file
<mat>/workspace/.metadata/.log ?

Krum

Re: Using pre-indexed dump [message #5669 is a reply to message #5479]

Fri, 17 October 2008 18:25

Eclipse User

Interestingly, my <mat> directory tree does not contain a workspace,
metadata or a .log directory or file.

In the end, I ended up running MAT on the 64-bit machine and using X to
show the GUI on my 32-bit machine. Even then, I had to add some parameters
to prevent OutOfMemoryError and "GC limit exceeded" errors:

mat64/MemoryAnalyzer -vmargs -Xmx2000m -XX:-UseGCOverheadLimit

Previous Topic:	Configuring JRE for running MAT
Next Topic:	Interned strings compete in the PermGen space.......

Goto Forum:

-=] Back to Top [=-

Current Time: Fri Apr 25 06:28:55 EDT 2025

.:: Contact :: Home ::.

Breadcrumbs

Sign up to our Newsletter