We should apply Debian's language shootout and adapt it to use J9. My initial testing by running the C# and Java nbody tests show that Mono is twice as fast as J9 (in AIX on POWER7, J9 8.0 64-bit, assuming both runtimes' defaults) in it, despite not having optimizations for CPUs post-POWER5. We should use the normal benchmark runner, but it's not very well documented; we'll have to play with it further some day.
If we can gather the data, Kyte from upstream is very interested in seeing our data and what tools we can use in AIX or PASE for profiling.
We should apply Debian's language shootout and adapt it to use J9. My initial testing by running the C# and Java
nbodytests show that Mono is twice as fast as J9 (in AIX on POWER7, J9 8.0 64-bit, assuming both runtimes' defaults) in it, despite not having optimizations for CPUs post-POWER5. We should use the normal benchmark runner, but it's not very well documented; we'll have to play with it further some day.If we can gather the data, Kyte from upstream is very interested in seeing our data and what tools we can use in AIX or PASE for profiling.