[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: instruction cache throttling



Joerg Sommer <joerg@alea.gnuu.de> writes:
> Joerg Sommer <joerg@alea.gnuu.de> wrote:
>> Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
>>> On Sat, 2005-05-14 at 21:11 +0000, Joerg Sommer wrote:
>>>> Joerg Sommer <joerg@alea.gnuu.de> wrote:
>>>> > Here is the hack http://www.minet.uni-jena.de/~joergs/ictc/. I don't know
>>>> > if its correct, but it doesn't make the kernel oops.
>>>> 
>>>> I used this module today and I don't see any difference to the life
>>>> without. I use 254 because 255 cause very high cpu load.
>>>
>>> Did you see a significant difference in battery consumption with
>>> different ictc values ?
>>
>> No. I think it doesn't work.
>
> I took the false values. It is not only putting the value into the
> register. To enable ictc bit one of the register must be enabled.
> Unfortunately I tried powers of two. :-( Now with 15 (divider = 10) it
> works and the system behaves like a slower system. Playing a mp3 with
> xmms consumes nearly 50 percent of cpu.
>
> But I observed a higher power consumption. I don't know why.

I think a higher consumption could be explained, but I haven't looked at it in
detail yet. It depends on what Linux does with the CPU when there are no tasks
running.

If the idle state uses less power than the running state, then it is probably
better to run the instructions as quickly as possible, and then spend as much
time as possible in the idle state. If this is the case (which could be
verified by looking at kernel source code and the device datasheet), then ICTC
is only going to reduce power consumption for loads which will be using 100%
of CPU at high speed. But this application is not particularly useful for a
desktop machine.

cheers,

John

Attachment: pgpYjPc7wumKU.pgp
Description: PGP signature


Reply to: