[Devel] Re: RSS controller v2 Test results (lmbench )

[Balbir Singh]

Rik van Riel wrote:
> Balbir Singh wrote:
> 
>> A meaningful container size does not hamper performance. I am in the
>> process
>> of getting more results (with varying container sizes). Please let me
>> know
>> what you think of the results? Would you like to see different
>> benchmarks/
>> tests/configuration results?
> 
> AIM7 results might be interesting, especially when run to crossover.
> 

I'll try and get hold of AIM7, I have some AIM9 results (please
see the attachment, since the results overflow 80 columns, I've
attached them).

> OTOH, AIM7 can make the current VM explode spectacularly :)
> 
> I saw it swap out 1.4GB of memory in one run, on my 2GB memory test
> system.  That's right, it swapped out almost 75% of memory.
> 

This would make a good test case for the RSS and the unmapped page
cache controller. Thanks for bringing it to my attention.

> Presumably all the AIM7 processes got stuck in the pageout code
> simultaneously and all decided they needed to swap some pages out.
> However, the shell got stuck too so I could not get sysrq output
> on time.
> 

oops! I wonder if AIM7 creates too many processes and exhausts all
memory. I've seen a case where during an upgrade of my tetex on my
laptop, the setup process failed and continued to fork processes
filling up 4GB of swap.

> I am trying out a little VM patch to fix that now, carefully watching
> vmstat output.  Should be fun...
> 

VM debugging is always fun!

-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: aim9
URL: <http://lists.openvz.org/pipermail/devel/attachments/20070518/9bb3b4fa/attachment-0001.ksh>