Re: Limiting the amount of memory an application can allocate

[Rigoberto Corujo <rcorujo yahoo com>]

Thank you for the clarification.

> I was going to test this, but at least on F7 x86_64,
ulimit -m doesn't actually have any effect.

This is an interesting comment.  I was getting hung up
on trying to figure out why "ulimit -m" didn't seem to
do anything, but I think you've just confirmed that it
doesn't do anything.  Seems like "ulimit -v" is what I
should be focusing on.

Thanks to everyone who responded.

Rigoberto Corujo

--- John Haxby <jch scalix com> wrote:

> Rigoberto Corujo wrote:
> >
> >> The resident set size, rss, is the maximum amount
> of
> >> virtual memory resident in real memory.
> >>     
> >
> > Would you be able to explain what "maximum amount
> of
> > virtual memory resident in real memory" means?
> >
> >   
> Hmm.   Not had to explain this one for a while ...
> 
> You know what virtual memory is, right?   Each
> process on the system 
> believes that is has all the memory allocated to it
> that it's asked 
> for.   If you do "malloc(1024*1024*1024)" then as
> far as the process is 
> concerned it has 1GB memory allocated to it. 
> However, unless you start 
> trying to use that allocated memory, the kernel
> doesn't actually map its 
> pages to real memory.
> 
> The number of pages currently mapped to real memory
> is called the 
> resident set size, rss.  The resident set size
> depends on several 
> things: the virtual size of the process (duh), the
> working set size of 
> the process, the amount of memory in the machine and
> how much other 
> processes want to use that memory.
> 
> If you artificially limit the resident set size to
> something less than 
> the working set size of the process then you'll just
> cause lots of 
> paging and the process will run very very slowly. 
> In fact the whole 
> system will tend to run slowly because the disk will
> be thrashing its 
> silly self to death.
> 
> I was going to test this, but at least on F7 x86_64,
> ulimit -m doesn't 
> actually have any effect.   That's probably a good
> thing.  I do use 
> ulimit -v though -- I use xlhtml to convert excel
> spreadsheets to ascii, 
> but occasionally it trips over a bad spreadsheet and
> goes mad allocating 
> memory so it's limited to 100Mb to stop it grabbing
> all the allocatable 
> memory and forcing the oomkiller to step in.
> 
> jch
> 
> jch
> 
> --
> nahant-list mailing list
> nahant-list redhat com
> https://www.redhat.com/mailman/listinfo/nahant-list
> 



      ____________________________________________________________________________________
Fussy? Opinionated? Impossible to please? Perfect.  Join Yahoo!'s user panel and lay it on us. http://surveylink.yahoo.com/gmrs/yahoo_panel_invite.asp?a=7