Alan Cox <alan@xxxxxxxxxx> writes:
> On Mon, Jan 17, 2005 at 03:30:50PM +0100, Nicolas Mailhot wrote:
>> bytecode. I don't think there is anyone made the argument than gcj +
>> native is faster than commercial jvm + bytecode.
>
> Do the benchmarks on a real world test set or for that matter just compare
> the start up time of a proprietary commercial jv eclipse and a compiled one.
>
> The compiled stuff starts much faster and consumers astronomically less
> memory and memory bandwidth when running. There are also some nice java
> bindings for Gnome which are useful when you need to run the same core code
> with a different UI on your phone and Linux box.
I don't think it's as clear cut as that. App startup time matters
for, say, CLIs (but who would make a java cli app... well, except
ant, heh) and for GUI apps (which starts to matter a little more) but
not so much about some of the more interesting Java areas --
serverside. I think at best the jury is still out on how fast gcj is
vs a hotspot-optimizing jvm for "real world" use.
It would be interesting to see some benchmarks comparing real-world
workloads and examples. I don't know of any off hand, though, that
included gcj.
Chip
--
Chip Turner cturner@xxxxxxxxxx
Red Hat, Inc.
