Re: [Results] [RFC PATCH v4 00/40] mm: Memory Power Management

[Srivatsa S. Bhat]

On 09/26/2013 06:28 PM, Srivatsa S. Bhat wrote:
> On 09/26/2013 05:10 AM, Andrew Morton wrote:
>> On Thu, 26 Sep 2013 04:56:32 +0530 "Srivatsa S. Bhat" <srivatsa.bhat@xxxxxxxxxxxxxxxxxx> wrote:
>>
>>> Experimental Results:
>>> ====================
>>>
>>> Test setup:
>>> ----------
>>>
>>> x86 Sandybridge dual-socket quad core HT-enabled machine, with 128GB RAM.
>>> Memory Region size = 512MB.
>>
>> Yes, but how much power was saved ;)
>>
> 
> I don't have those numbers yet, but I'll be able to get them going forward.
> 

Hi,

I performed experiments on an IBM POWER 7 machine and got actual power-savings
numbers (upto 2.6% of total system power) from this patchset. I presented them
at the Kernel Summit but forgot to post them on LKML. So here they are:

Hardware-setup:
--------------
IBM POWER 7 machine: 4 socket (NUMA), 32 cores, 128GB RAM

 - 4 NUMA nodes with 32 GB RAM each
 - Booted with numa=fake=1 and treated them as 4 memory regions


Software setup:
--------------
Workload: Run modified ebizzy for half an hour, which allocates and frees large
quantities of memory frequently. The modified ebizzy touches every allocated
page a number of times (4 times) before freeing it up. This ensures that
allocating a page in the "wrong" memory region makes it very costly in terms
of power-savings, since every allocated page is accessed before getting
freed (and accesses cause energy consumption). Thus, with this modified
benchmark, sub-optimal MM decisions (in terms of memory power-savings) get
magnified and hence become noticeable.


Power-savings compared to mainline (3.12-rc4):
---------------------------------------------
With this patchset applied, the average power of the system reduced by 2.6%
compared to the mainline kernel during the benchmark run. The total system
power is an excellent metric for such evaluations, since it brings out the
overall power-efficiency of the patchset. (IOW, if the patchset shoots up the
CPU or disk power-consumption while causing memory power savings, then the
total system power will not show much difference). So these numbers indicate
that the patchset performs quite well in reducing the power-consumption of
the system as a whole.

This is not the most ideal hardware configuration to test on, since I had
only 4 memory regions to play with, but this gives a good initial indication
of the kind of power savings that can be achieved with this patchset.

I am expecting the same patchset to give us power-savings of upto 5% of the
total system power on a newer prototype hardware that I have (since it has
more memory regions and lower base power consumption).


Regards,
Srivatsa S. Bhat

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/