Re: [Gnu-arch-users] Nit

[Charles Duffy]

On Mon, 2003-10-20 at 03:23, Thomas Zander wrote:
> I think you want to do some profiling using recent JVMs since your problem 
> not really an issue since 1.4.x

Thomas, I'll agree that general-purpose (post-startup) performance is
not really an issue since 1.4.x -- but *JVM startup time* is another
matter.

Since you ask for real benchmarks... hell, you're mostly right. The time
for a simple 1-class HelloWorld app using Sun's JVM is now only about
40% slower than the same trivial "program" compiled using gcj, and the
difference isn't (on my 2GHz P4m laptop) exactly a perceptable one.

Now, I understand that the difference can get much larger under the
right circumstances -- folks using the gcj-compiled Eclipse build, for
instance, have reported it to be much faster than the traditional one --
but nonetheless, in assuming a constant 1-second-plus JVM startup time
cost, I was indeed in error.


(That said, I still would end up compiling tla via gcj for my own use --
particularly if it's being repeatedly called by my editor, and thus
restarting itself frequently, even small increments of time can add up).