time cat /proc/*/statm ?

time cat /proc/*/statm ?

[u1_amd64]

Hi,

Is it reasonable for a 64bit dual cpu to take 5+ seconds of processing to
cat /proc/*/statm when there is hardly more than 1gb of actual memory
space used by processes (the rest being filesystem cache)?

This makes top or anything else that uses statm, unusable.

I don't observe this behavior on a 4gb 32bit machine, with the same
working set of mysqld processes, and 3gb of filesystem cache.

# free
             total       used       free     shared    buffers     cached
Mem:      16278356   16264484      13872          0      85400   14819932
-/+ buffers/cache:    1359152   14919204

# ps -eda|wc
 216     866    6751

# time cat /proc/*/statm
:
:
:
real    0m5.740s
user    0m0.003s
sys     0m5.521s

# uname -a

Linux silver 2.4.21-151-smp #22 SMP Mon Jan 5 21:31:07 PST 2004 x86_64 x86_64 x86_64 GNU/Linux


thanks!!

Re: time cat /proc/*/statm ?

[Roger Luethi]

On Thu, 08 Jan 2004 13:54:58 -0500, u1_amd64@dslr.net wrote:
> Is it reasonable for a 64bit dual cpu to take 5+ seconds of processing to
> cat /proc/*/statm when there is hardly more than 1gb of actual memory
> space used by processes (the rest being filesystem cache)?
> 
> This makes top or anything else that uses statm, unusable.

Why does top still read /proc/*/statm anyway? It's not as if top actually
ever used that information (the top I looked at at the time, that is). I
submitted a patch a few months ago to remove statm because it is a)
broken and b) redundant. The message containing detailed reasoning
should be in the linux-mm archives.

Roger

Re: time cat /proc/*/statm ?

[Willy Tarreau]

On Thu, Jan 08, 2004 at 01:54:58PM -0500, u1_amd64@dslr.net wrote:
> Is it reasonable for a 64bit dual cpu to take 5+ seconds of processing to
> cat /proc/*/statm when there is hardly more than 1gb of actual memory
> space used by processes (the rest being filesystem cache)?

Here on alpha, it takes 0.023s (0.011 sys) for 58 processes on 2.4.23.
Perhaps there's something specific to x86_64 ?

Willy

> # free
>              total       used       free     shared    buffers     cached
> Mem:      16278356   16264484      13872          0      85400   14819932
> -/+ buffers/cache:    1359152   14919204
> 
> # ps -eda|wc
>  216     866    6751
> 
> # time cat /proc/*/statm
> :
> :
> :
> real    0m5.740s
> user    0m0.003s
> sys     0m5.521s
> 
> # uname -a
> 
> Linux silver 2.4.21-151-smp #22 SMP Mon Jan 5 21:31:07 PST 2004 x86_64 x86_64 x86_64 GNU/Linux
> 
> 
> thanks!!
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: time cat /proc/*/statm ?

[Albert Cahalan]

Roger Luethi writes:

> Why does top still read /proc/*/statm anyway?
> It's not as if top actually ever used that
> information (the top I looked at at the time,
> that is). I submitted a patch a few months ago
> to remove statm because it is a) broken and
> b) redundant. The message containing detailed
> reasoning should be in the linux-mm archives.

top (from procps-3.1.14 at least) only reads
the /proc/*/statm files when you activate
columns that need the statm data. Those are:

%MEM   Memory usage (RES) 
VIRT   Virtual Image (kb) 
SWAP   Swapped size (kb) 
RES    Resident size (kb) 
CODE   Code size (kb) 
DATA   Data+Stack size (kb) 
SHR    Shared Mem size (kb)
nDRT   Dirty Pages count 

By default, some of those get displayed.
The default includes: VIRT RES SHR %MEM

To get rid of those, type this:
   f  o  q  t  n  enter  W
(the "W" writes out a config file)

In case your top is too old for this, you
can go to http://procps.sf.net/ for a new one.