Holger Hoffstätte wrote:
> No, we're talking (on my machine) 15 milliseconds vs. ~2 seconds - the
> interface test takes noticeably longer than the other two. If that is not
> significant, I don't know what is.
cool down ... :-)
That's exactly what I said. 2 seconds / 100 million is
20 nanos, right?
An object allocation as well as a synchronized lock is nearly a micro.
> Of course ... a simple allocation will totally throw this off
You name it :-)
> Crap is not excusable just because you don't see it.
That's where you are wrong. I tried this example,
and yes, I see a factor ~30 for sun-jdk13 in server mode.
But what does it mean?
It means that this VM is able to do the direct call in
30ms / 100 million = 0.3 nanoseconds - that's faster
than the clockspeed!
And you are saying everything else is crap? - come on!
That's nice research but totally irrelavant in practice.
The cpu-bottleneck in Cayenne is nothing else but
object allocation.
> The differences that can be achieved by microtuning
> can be _astounding_ and are usually horribly
> underestimated, even when the obvious strategies
> (algorithm, hardware..) have been applied.
I agree you should have the microtuning parameters
in mind, and this is 20 nanos for an interface call
and 500 nanos for an object allocation.
That's all I wanted to say.
schoene Gruesse,
Arndt
This archive was generated by hypermail 2.0.0 : Sat Mar 22 2003 - 12:43:59 EST