[erlang-questions] HiPE performance gain (Was: Re: [erlang-questions] Erlang static linked interpreter)

[Matthias Lang]

MML wrote:
> > I've assumed that what happens is that each new Erlang release is
> > tuned for increasingly faster CPUs, while our hardware doesn't change
> > as often. But I'm just guessing.

JL wrote:
> In my experience, the only way to combat performance regressions like
> these is to plot the development branch against the time to complete

Did you interpret my comment to mean "The OTP guys aren't noticing 3%
preformance regressions which hit everyone. They're not noticing
because they get new, faster machines every year"? That's not what I
meant.

What I meant was "In 2004, a typical server CPU was clocked (say) 2x
faster than an Au1000 and had (say) 8x as much cache. In 2010, the
ratios are more like 5x and 64x. This increased difference might* have
made it more likely for a given change to the VM to result in
increased performance on a current server CPU while also causing a
decrease on an Au1000."

I have no reason to believe that performance gets worse from release
to release when running on normal (e.g. x86 server) hardware.

Matt

 * i.e. wild speculation, discussion would be interesting if someone
   has concrete examples or, better still, actual measurements.