memory use question
Daniel Miller
It's complicated, though, by difficulty interpreting the output from
/proc/meminfo. I would normally expect that Shared+Buffers+Cached would
more or less equal Used. However, in a busy system, with alot of
continuous network traffic going on, we often see that S+B+C MUCH_LESS_THAN
Used... for example, Used=139MB, S+B+C=55MB (in a 196MB machine).
Is the missing memory actually leaked memory?? Or is it just tied up in
kernel buffers that aren't reflected in /proc/meminfo??
As a separate question, is there something that I can look in the system
that WILL enumerate all memory usage in some consistent manner? I tried
summing up all /proc/PID/statm numbers, but they neither account for all
the used memory, nor do the resulting numbers correlate directly with any
of the values in /proc/meminfo ...
-----------== Posted via Newsfeed.Com - Uncensored Usenet News ==----------
http://www.newsfeed.com The #1 Newsgroup Service in the World!
-----= Over 100,000 Newsgroups - Unlimited Fast Downloads - 19 Servers =-----
Kasper Dupont
>
> We're trying to figure out if we've got a memory leak in our system here.
> It's complicated, though, by difficulty interpreting the output from
> /proc/meminfo. I would normally expect that Shared+Buffers+Cached would
> more or less equal Used. However, in a busy system, with alot of
> continuous network traffic going on, we often see that S+B+C MUCH_LESS_THAN
> Used... for example, Used=139MB, S+B+C=55MB (in a 196MB machine).
>
> Is the missing memory actually leaked memory?? Or is it just tied up in
> kernel buffers that aren't reflected in /proc/meminfo??
Does this script help you?
http://www.daimi.au.dk/~kasperd/linux_kernel/slabs
>
> As a separate question, is there something that I can look in the system
> that WILL enumerate all memory usage in some consistent manner? I tried
> summing up all /proc/PID/statm numbers, but they neither account for all
> the used memory, nor do the resulting numbers correlate directly with any
> of the values in /proc/meminfo ...
I don't think it is possible to get correct accounting for all memory.
But maybe you can write a kernel module to scan the page list and
tell you what each page is used for. I don't know if you can tell
exactly what each page is used for.
--
Kasper Dupont -- der bruger for meget tid på usenet.
For sending spam use mailto:aaa...@daimi.au.dk
for(_=52;_;(_%5)||(_/=5),(_%5)&&(_-=2))putchar(_);
Daniel Miller
news:3E23A148...@daimi.au.dk:
> Daniel Miller wrote:
>>
>> We're trying to figure out if we've got a memory leak in our system
>> here. It's complicated, though, by difficulty interpreting the output
>> from /proc/meminfo. I would normally expect that
>> Shared+Buffers+Cached would more or less equal Used. However, in a
>> busy system, with alot of continuous network traffic going on, we
>> often see that S+B+C MUCH_LESS_THAN Used... for example, Used=139MB,
>> S+B+C=55MB (in a 196MB machine).
>>
>> Is the missing memory actually leaked memory?? Or is it just tied up
>> in kernel buffers that aren't reflected in /proc/meminfo??
>
> Does this script help you?
> http://www.daimi.au.dk/~kasperd/linux_kernel/slabs
Unfortunately, I'm using kernel 2.2, which has slabinfo 1.0; it doesn't
have as much info as later versions... it may provide *some* useful info,
tho... thanks!!
>
>>
>> As a separate question, is there something that I can look in the
>> system that WILL enumerate all memory usage in some consistent
>> manner? I tried summing up all /proc/PID/statm numbers, but they
>> neither account for all the used memory, nor do the resulting numbers
>> correlate directly with any of the values in /proc/meminfo ...
>
> I don't think it is possible to get correct accounting for all memory.
> But maybe you can write a kernel module to scan the page list and
> tell you what each page is used for. I don't know if you can tell
> exactly what each page is used for.
>
-----------== Posted via Newsfeed.Com - Uncensored Usenet News ==----------
Kasper Dupont
> > http://www.daimi.au.dk/~kasperd/linux_kernel/slabs
>
> Unfortunately, I'm using kernel 2.2, which has slabinfo 1.0; it doesn't
> have as much info as later versions... it may provide *some* useful info,
> tho... thanks!!
A large usage of the buffer_head and dentry_cache is normal. The
two columns with numbers are the number of used and allocated
entries. For most slabs there are allocated one page at a time,
so if you can find out in how large steps the allocated field
grows, you can probably expect this number of elements to fit in
a single page. Otherwise you will have to start looking in the
source to find the size. For the size-%d fields the number is
the size of each element in bytes.