lua-users home
lua-l archive

[Date Prev][Date Next][Thread Prev][Thread Next] [Date Index] [Thread Index]

I was not trying to say that JET is the fastest JVM around. It would
be unethical and false.

My point was: different JVMs have different weak and strong sides, so
they can show different performance boost for different applications.
To make things more complicated: for language like Java you should
consider not only optimization differences (e.g. approaches to loop
versioning, to global optimizations), code generation differences
(e.g. SSE2 vs. FPU, or different implementations of operations with
"long" type), but also differences in runtime support (different
approach to GC, special means to optimize reflection etc.).

You just can't say that "HS performance" is equal to "Java performace".

Nevertheless all this is irrelevant to the whole topic of Lua JIT.
Sorry for the disturbance, I should have waited for the actual sample
for benchmarking before saying anything.

/* BTW, Krause uses not arithmetic, but _geometric_ mean. Also it
measures ratio not between HS and JET, but between X and the fastest
competitor for each test.
So "unduly large" is not very large. I did some finger-math to exclude
it: HS score will be ~1.35226.
But this is also irrelevant, we are not on Java lovers list ;-)
If you would like to continue this argument, feel free to contact me directly.


On Mon, Nov 2, 2009 at 10:32 PM, Isaac Gouy <> wrote:
> Vyacheslav Egorov <mister.aleph <at>> writes:
> -snip-
>> Also there were some benchmarks[1] showing that HotSpot-Server is not
>> the fastest JVM around. It might be interesting to test the very same
>> number crunching app on JRockit or Excelsior JET.
> -snip-
>> [1]; methodology of the
>> benchmarking was not flawless, so results should be interpreted with
>> caution.
> -snip-
> Only one (fannkuch) of the four programs that Stefan Krause borrowed from the
> benchmarks game shows Excelsior JET much faster than HotSpot.
> The fifth and final program (himeno) shows Excelsior JET 4x faster and that
> (outlier?) has an unduly large effect on the mean Stefan Krause shows.
> Let's be clear - the benchmarks game isn't flawless either! ;-)
> However, the benchmarks game has been open to incremental improvements over the
> last 4 or 5 years - in fact, median (more resistant to outliers) and quartiles
> are shown instead of a mean because someone came and made a persuasive argument.
> And of course programmers contribute better programs - the Java fannkuch program
> Stefan Krause borrowed is no longer shown, it's been replaced by 2 faster Java
> programs.
> No doubt "someone" will contribute better LuaJIT programs :-)