| From: | Rudi Chiarito |
| Date: | 22 Sep 2000 at 18:15:32 |
| Subject: | Re: AMIOPEN: XML and Java, a uncharted territory? |
On Fri, Sep 22, 2000 at 05:25:51PM +0100, John Niclasen wrote:
> So garbage collection can be done in the background, but
> you need some clever minds to implement it, and it seems,
> we got that. :)
Several months ago I tested Tao's JVM with a program by Thomas Rokicki
(AmigaTeX [for many years, the best implementation of TeX, by far],
dvips, Maple for Amiga, etc.). He wrote it to reproduce a problem with
Sun's collector, i.e. the fact that the load of the garbage collector was
not evenly distributed in time: its CPU usage would stay quite low for
some time, then it would get quite high for a short while, then low
again, etc. (you can find more details at the JDC, it's one of the long
standing bugs). That's quite unfortunate for embedded and/or realtime
applications, as it can introduce "hiccups". The problem seems to be
still there with recent versions of Sun's JVM, although things are
better, because the collector is faster and thus "hiccups" have become
harder to notice.
Anyway, I was delighted to see that the program ran just fine with Tao's
JEngine, as the latency was pretty much constant. No wonder, considering
it is the JVM that Scott McNealy used to demonstrate Java on a phone
(it's funny that it uses not even one line of Sun's code ;) ).