Cog and Batteries?

I have a perhaps naive question: what's the expected impact of the JIT on battery life? Seems like the extra number crunching might have an impact. Regardless of whether or not people want us generating native code on mobile devices, does it generally make more sense to use the stack VM on mobile devices?

FWIW, I think 10-15% better performance is still a f@(&!?g big deal on dog slow devices! Oh, this stuff just makes my whole year fantastic.

Of course, the question may be a bit premature since we don't have a JIT for ARM yet:)

On 22.06.2010, at 23:15, Casey Ransberger wrote:

> I have a perhaps naive question: what's the expected impact of the JIT on battery life? Seems like the extra number crunching might have an impact. Regardless of whether or not people want us generating native code on mobile devices, does it generally make more sense to use the stack VM on mobile devices?

The key to good battery live is to run as fast as possible for the shortest amount of time possible, then go back to sleep. So yes, Cog would help for battery IMHO - unless it seduces you to also do more.

> FWIW, I think 10-15% better performance is still a f@(&!?g big deal on dog slow devices! Oh, this stuff just makes my whole year fantastic.
>
> Of course, the question may be a bit premature since we don't have a JIT for ARM yet:)

Would be awesome if someone would tackle that.

- Bert -

On 6/22/2010 2:23 PM, Bert Freudenberg wrote:
>> Of course, the question may be a bit premature since we don't have a JIT for ARM yet:)
>
> Would be awesome if someone would tackle that.

Hm .... I don't have the exact values at hand, but from memory our speed
improvements were roughly:

        SqueakVM -> StackVM: 1.5x faster
        StackVM  -> JitVM:     2x faster

So there's probably a good speedup to be had for mobile devices from
*just* using the StackVM (StackInterpreter) instead of the regular
Interpreter.

Cheers,
   - Andreas