lkml.org 
[lkml]   [2007]   [Jun]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 0/8] RSS controller based on process containers (v3.1)
Herbert Poetzl wrote:
> On Mon, Jun 04, 2007 at 05:25:25PM +0400, Pavel Emelianov wrote:
>> Adds RSS accounting and control within a container.
>>
>> Changes from v3
>> - comments across the code
>> - git-bisect safe split
>> - lost places to move the page between active/inactive lists
>>
>> Ported above Paul's containers V10 with fixes from Balbir.
>>
>> RSS container includes the per-container RSS accounting
>> and reclamation, and out-of-memory killer.
>>
>>
>> Each mapped page has an owning container and is linked into its
>> LRU lists just like in the global LRU ones. The owner of the page
>> is the container that touched the page first.
>
>> As long as the page stays mapped it holds the container, is accounted
>> into its usage and lives in its LRU list. When page is unmapped for
>> the last time it releases the container.
>
>> The RSS usage is exactly the number of pages in its booth LRU lists,
>> i.e. the nu,ber of pages used by this container.
>
> so there could be two guests, unified (i.e. sharing
> most of the files as hardlinks), where the first one
> holds 80% of the resulting pages, and the second one
> 20%, and thus shows much lower 'RSS' usage as the
> other one, although it is running the very same
> processes and providing identical services?

Herbert!!! Where have you been so long?

You must have missed that we've decided not to account pages
sharing right now, but start with that model. Later we'll make
sharing accountable.

( There's something bad with my memory. I have a feeling that I
have already told that... many times... )

>> When this usage exceeds the limit set some pages are reclaimed from
>> the owning container. In case no reclamation possible the OOM killer
>> starts thinning out the container.
>
> so the system (physical machine) starts reclaiming
> and probably swapping even when there is no need
> to do so?

Good catch! The system will start reclaiming right when the
container hits the limit to expend its IO bandwidth. Not some
other's one that hit the global limit due to some bad container
was allowed to go above it.

> e.g. a system with a single guest, limited to 10k
> pages, with a working set of 15k pages in different
> apps would continuously swap (trash?) on an otherwise
> unused (100k+ pages) system?
>
>> Thus the container behaves like a standalone machine - when it runs
>> out of resources, it tries to reclaim some pages, and if it doesn't
>> succeed, kills some task.
>
> is that really what we want?

A kind of ;)

> I think we can do _better_ than a standalone machine
> and in many cases we really should ...

That's it! You are right - this is our ultimate goal. And we
plan to get there step by step. And we will appreciate your
patches fixing BUGS, improving the performance, extending the
functionality, etc.

> best,
> Herbert

Thanks for your attention,
Pavel

>> Signed-off-by: Pavel Emelianov <xemul@openvz.org>
>>
>> The testing scenario may look like this:
>>
>> 1. Prepare the containers
>> # mkdir -p /containers/rss
>> # mount -t container none /containers/rss -o rss
>>
>> 2. Make the new group and move bash into it
>> # mkdir /containers/rss/0
>> # echo $$ > /containers/rss/0/tasks
>>
>> Since now we're in the 0 container.
>> We can alter the RSS limit
>> # echo -n 6000 > /containers/rss/0/rss_limit
>>
>> We can check the usage
>> # cat /containers/rss/0/rss_usage
>> 25
>>
>> And do other stuff. To check the reclamation to work we need a
>> simple program that touches many pages of memory, like this:
>>
>> #include <stdio.h>
>> #include <unistd.h>
>> #include <sys/mman.h>
>> #include <fcntl.h>
>>
>> #ifndef PGSIZE
>> #define PGSIZE 4096
>> #endif
>>
>> int main(int argc, char **argv)
>> {
>> unsigned long pages;
>> int i;
>> char *mem;
>>
>> if (argc < 2) {
>> printf("Usage: %s <number_of_pages>\n", argv[0]);
>> return 1;
>> }
>>
>> pages = strtol(argv[1], &mem, 10);
>> if (*mem != '\0') {
>> printf("Bad number %s\n", argv[1]);
>> return 1;
>> }
>>
>> mem = mmap(NULL, pages * PGSIZE, PROT_READ | PROT_WRITE,
>> MAP_PRIVATE | MAP_ANON, 0, 0);
>> if (mem == MAP_FAILED) {
>> perror("map");
>> return 2;
>> }
>>
>> for (i = 0; i < pages; i++)
>> mem[i * PGSIZE] = 0;
>>
>> printf("OK\n");
>> return 0;
>> }
>> _______________________________________________
>> Containers mailing list
>> Containers@lists.linux-foundation.org
>> https://lists.linux-foundation.org/mailman/listinfo/containers
>

-
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/

\
 
 \ /
  Last update: 2007-06-08 14:39    [W:0.462 / U:0.468 seconds]
©2003-2011 Jasper Spaans. Advertise on this site