[erlang-questions] clueless performance question

[Christian S]

On Tue, Jun 10, 2008 at 10:27 PM, Raoul Duke <raould@REDACTED> wrote:
> Erlang excels at concurrency and distribution. What aspects of the
> current Erlang VMs (HiPE and non-HiPE) make them possibly slower for
> serial computation vs. some other well-known VMs? If somebody wanted
> to work on tweaking the serial computation part, what might be the top
> things to examine (e.g. from personal experience, or from profiler
> usage)? This could be anything, like doing work on VM inner loops, or
> adding some more automatic optimization abilities.

Get the papers produced by the hipe group to see what they have done
to increase the sequential performance. It's an interesting read,
exposing what one have to win by using hipe.

I would find research into what can be won by doing JIT interesting,
things like avoiding external-call lookups by compiling it into a
direct call (constraint it so it needs to be re-done when modules are
replaced). I imagine these small book-keeping costs in dynamic
languages sum up, if not else, there would be no reason for java jits
to do it. :)

Of course, for the forseeable future, the winner in raw computation
will be those that know how to distribute the computation to make best
use of many simultaneous processing units.


PS.
Possibly nobody in research wants to disclose their current work
before they can get it into a paper to publish. It is how funded
research tend to work.