Re: [PATCH 2.6.14-rc2] fix incorrect mm->hiwater_vm and mm->hiwater_rss

[Jay Lan <jlan@engr.sgi.com>]

Jay Lan wrote:
> Hugh Dickins wrote:
> 
>>
>>
>> See comment in fs/proc/task_mmu.c for the principle.  Could maintain
>> hiwater_vm straightforwardly, but I think it's easier to remember if
>> we handle them both in the same way.
>>
>> I did look into doing the total_vm increment and calling vm_stat_account
>> in insert_vm_struct, but concluded it solved no particular problem, and
>> raised some questions (where architectures, notably ia64, have special
>> vmas which they may have good reason to leave out of total_vm).
>>
>> I haven't cross-checked the mm_struct cacheline rearrangement yet,
>> it looks plausible, but could easily turn out to straddle boundaries.
>>
>> Christoph, Frank, Jay: does this patch look like it fits your needs?
> 
> 
> I am building a kernel with your patch and am going to run some test
> to compare the statistics.

My testing showed the same number on hiwater_vm, but hiwater_rss from
Hugh's version was consistently ~1.5% lower. Where was the loss?

The fact that i have consistent hiwater_vm and hiwater_rss
in a few hundreds of processes suggests that that test may not
be a good test for comparing hiwater_vm and hiwater_rss.

I guess it allocates same amount of memory up front in every sub-tests
processes and never get over it. However, it also showed the way Hugh
did hiwater_rss in new code missed something.

Tomorrow i will be very busy at my work. I will be back on this
next Monday.

Thanks,
  - jay

> 
> Thanks,
>  - jay
> 
>>
> -
> 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/