[prev in list] [next in list] [prev in thread] [next in thread] 

List:       oprofile-list
Subject:    [ oprofile-Bugs-512417 ] Memory usage
From:       noreply () sourceforge ! net
Date:       2002-02-03 18:22:05
[Download RAW message or body]

Bugs item #512417, was opened at 2002-02-03 08:45
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=512417&group_id=16191

Category: None
Group: None
>Status: Closed
>Resolution: Invalid
Priority: 5
Submitted By: Yaakov Yaari (yaari)
>Assigned to: John Levon (movement)
Summary: Memory usage

Initial Comment:
This is most probably not a bug but rather my 
ignorance.
I noticed that as oprofiled runs, each time op_dump is 
issued, oprofiled' allocated memory , as seen by 
`top', gives a jump. This way, the allocated memory  
quickly grows to 25% of the total (I have a small 64M 
system). 
By instrumenting xmalloc/calloc/realloc/free I found 
few potential leaks, which are pretty small (like 
opd_proc.c:623, in opd_add_image()). The total leakage 
after few moments was ~300K. However these do not 
explain the above large percenatge (25% and growing). 

- oprofile version: 0.0.8
- Linux version: 2.4.2
Thanks
Yaakov

----------------------------------------------------------------------

>Comment By: John Levon (movement)
Date: 2002-02-03 18:21

Message:
Logged In: YES 
user_id=53034

OK, I've checked some quick runs with dmalloc. All reports
were for stuff not de-allocated at shutdown. Whilst this
doesn't really matter, I've just committed some code to
free this stuff anyway.

A clarification: by swapped out I mean swapped back to the
sample file on disk.

----------------------------------------------------------------------

Comment By: John Levon (movement)
Date: 2002-02-03 09:50

Message:
Logged In: YES 
user_id=53034

oprofiled keeps mmap()s of each sample file for performance.
It is up to the VM to keep the page usage reasonable - unused
pages get swapped out. You are likely to run into performance
problems with only 64Mb in this case.

Do you see significant perf
ormance problems when oprofiled is
running ? What kernel version are you using ?

In my low-mem tests, overhead has still been reasonable and
it hasn't impacted the machine significantly.

As for malloc leaks, there may be some, we still need to do
a proper audit. But any leaks that may or may not be
present are bounded.

The crucial figure here is RSS not VSZ.

----------------------------------------------------------------------

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=116191&aid=512417&group_id=16191


[prev in list] [next in list] [prev in thread] [next in thread]