Hi,
I am doing some performance analysis on machines running AIX and I discovers
something "strange" for me:
How could you explain a RESident set size greater than the VIRtual set size
for a specific process???

Thanks for your help
Pascal

[ Post a follow-up to this message ]

In comp.unix.aix Pascal Perrot <perrot_pascal@emc.com> wrote:
PP> Hi,
PP> I am doing some performance analysis on machines running AIX and I disco
vers
PP> something "strange" for me:
PP> How could you explain a RESident set size greater than the VIRtual set s
ize
PP> for a specific process???

Does VSS contain SysV shared memory pages?

Regards,

Nicholas

--
+--------------------------------------------------------------------------+
| Bush's Flip Flops                                                        |
|     * Bush is against campaign finance reform; then he's for it.         |
|     * Bush is against a Homeland Security Department; then he's for it.  |
| See http://www.dailykos.com/section/bush_admin for the complete list.    |
+--------------------------------------------------------------------------+

[ Post a follow-up to this message ]

Yes there are some shared libs.

I already thought about this thing, but from my side I think that at least
"all the res area" could be paged, so the same virtual area is already
reserved on disk in case on paging or swapping.
Don't forget this my question is about one process (not the whole machine).

You are right, frequently the shared part is counted n times (n=number of
processes using the shared area).

Regards,
Pascal

"Nicholas Dronen" <ndronen@io.frii.com> a écrit dans le message de
news:404dadd2$0$195$75868355@news.frii.net...
> In comp.unix.aix Pascal Perrot <perrot_pascal@emc.com> wrote:
> PP> Hi,
> PP> I am doing some performance analysis on machines running AIX and I
discovers
> PP> something "strange" for me:
> PP> How could you explain a RESident set size greater than the VIRtual set
size
> PP> for a specific process???
>
> Does VSS contain SysV shared memory pages?
>
> Regards,
>
> Nicholas
>
> --
>
+--------------------------------------------------------------------------+
> | Bush's Flip Flops
|
> |     * Bush is against campaign finance reform; then he's for it.
|
> |     * Bush is against a Homeland Security Department; then he's for it.
|
> | See http://www.dailykos.com/section/bush_admin for the complete list.
|
>
+--------------------------------------------------------------------------+

[ Post a follow-up to this message ]

Pascal Perrot wrote:
>
> I am doing some performance analysis on machines running AIX and I discove
rs
> something "strange" for me:
> How could you explain a RESident set size greater than the VIRtual set siz
e
> for a specific process???

AIX uses a lazy algorthym to assign disk space to pages.  Disk copies are
not mapped until it is actually needed to put the page onto disk.  If the
difference is large, the virtual size just might be measuring the disk
resident part instead of the total.

[ Post a follow-up to this message ]

In article <7960d3ee.0403090654.20a64e42@posting.google.com>,
dfreybur@yahoo.com (Doug Freyburger) wrote:

> Pascal Perrot wrote: 
>
> AIX uses a lazy algorthym to assign disk space to pages.  Disk copies are
> not mapped until it is actually needed to put the page onto disk.  If the
> difference is large, the virtual size just might be measuring the disk
> resident part instead of the total.

You have it backward -- "resident" means "resident in main memory".

The reason why resident size can be larger than virtual size is because
VSZ typically only counts the data and stack segments.  But portions of
the text segment also have to be in main memory for the program to run,
and it's possible for this to exceed the amount of data/stack memory.

--
Barry Margolin, barmar@alum.mit.edu
Arlington, MA
*** PLEASE post questions in newsgroups, not directly to me ***

[ Post a follow-up to this message ]

It's a guess.
Probably AIX does not show the total virtual size in vsz.
It might just show the executables text data bss and stack "segments"
but not later mapped objects (libaries files etc)
Thats what happens in NetBSD and OpenBSD with the same effect vsz < rss
and yes that's strange to me too.

I think I have read once it's the same with AIX.

(For NetBSD and OpenBSD thats not totally true as sbrk syscall
increase/decrease the vsz size but not file mappings by mmap)

Lars


Pascal Perrot wrote:

> Hi,
> I am doing some performance analysis on machines running AIX and I
> discovers something "strange" for me:
> How could you explain a RESident set size greater than the VIRtual set
> size for a specific process???
>
> Thanks for your help
> Pascal

[ Post a follow-up to this message ]