[omr-dev] Performance of SOM++ on OMR

[raffaello.giulietti@xxxxxxxx]

Hello,

in his video at the JVM Language Summit 2016
(https://www.youtube.com/watch?v=w5rcBiOHrB0), Mark Stoodley reports how
a one week implementation of a JIT for SOM++ allows to get a 3x-4x
performance improvement.

On the other side, another researcher (see the performance chart at
http://som-st.github.io/) reports a *100x-200x* performance improvement
over SOM++ with a Truffle/Graal implementation. The improvement has been
measured over a set of 4 standard benchmarks.

At first glance, thus, OMR results do not seem very impressive, but
perhaps it's like comparing apples and oranges. Something similar,
however, seems to hold for Ruby/OMR versus JRuby/Truffle.

So, why should somebody choose to adopt the OMR approach to language
runtimes rather than the Truffle/Graal alternative?

I know I might sound a little bit provocative, but this is intended to
stimulate hopefully technical and non religious discussions ;-)

Greetings
Raffaello