Hydra performance [was: Problem with HydraPing and FFI]

Igor Stasenko siguctua at gmail.com
Sun Feb 17 11:43:00 UTC 2008


On 17/02/2008, Klaus D. Witzel <klaus.witzel at cobss.com> wrote:
>
> Nah, what Gulik asked in the "Interrupt checking in HydraVM using event
> system" thread in squeak-vm. BTW: squeak-vm list does not show up all
> messages (for me), some I only see in NNTP's
> gmane.comp.lang.smalltalk.squeak.vm.devel ...
>

I don't expect too much speed loss with new refactorings. Instead, i'm
expecting some speedup :)
- there is no need to do checks for every N ms (or nanoseconds), since
interrupt to handle events now is possible only, when there are events
pending.
Now, the only (uniform) way to interrupt interpreter loop is by
posting event, notifying system that there is something needs to be
done beyond it.
And check for pending events is a simple read of pointer value (can't
be too much slower than decrementing interruptCheckCounter :)

> > It's an impartial drawback of objectification, there's little what we
> > can do about this.
>
> Yes, that's a tough one.
>
> > I didn't putted benchmarks in public, but i measured them.
> > - microBenchmarks shows about 5% bytecode crunching speed loss
> > - macroBenchmarks (there are 4 of different macro benchmarks which
> > Andreas gave me) showing about 10% speed loss on my box, interesting
> > that of them runs faster, without any reason :)
>
> Which ones: some as in "don't know which one"? Are they posted somewhere?
>
>
>


-- 
Best regards,
Igor Stasenko AKA sig.



More information about the Squeak-dev mailing list