The post I forward (from a java-related newsgroup) praises somehow
the speed of Allegro Lite.
Regards,
-- Xavier
<forwarded>
Path:
news.iddeo.es!news-1.retevision.es!195.22.205.162.MISMATCH!/news!nntp-cust.primenet.com!nntp.primenet.com!nntp.gctr.net!newsfeed.cwix.com!cyclone.columbus.rr.com!news.rr.com!cyclone-midwest.rr.com!typhoon.columbus.rr.com.POSTED!not-for-mail
From: "Bruce Tobin" <······@columbus.rr.com>
Newsgroups: comp.lang.java.advocacy,comp.lang.java.programmer
Subject: Java Performance Revisited.
Lines: 26
X-Priority: 3
X-MSMail-Priority: Normal
X-Newsreader: Microsoft Outlook Express 5.00.2919.6600
X-MimeOLE: Produced By Microsoft MimeOLE V5.00.2919.6600
Message-ID: <····················@typhoon.columbus.rr.com>
Date: Fri, 17 Mar 2000 02:20:08 GMT
X-Complaints-To: ·····@rr.com
X-Trace: typhoon.columbus.rr.com 953259608 24.26.152.25 (Thu, 16 Mar 2000 21:20:08 EST)
NNTP-Posting-Date: Thu, 16 Mar 2000 21:20:08 EST
Organization: Road Runner Columbus
Xref: news.iddeo.es comp.lang.java.advocacy:29095 comp.lang.java.programmer:81458
A couple of years ago I did a language performance comparison using a
benchmark based on the CN2 rule induction algorithm. The algorithm had a
mix of integer and floating point operations, a lot of operations on
variable-length sequences, and lots of dynamic memory allocation. The
results, at the time, showed that C++ had a significant lead over Java in
performance. The fastest time I coaxed out of a Java implementation
(Microsoft's J++ 6.0) was an average of around 6 seconds over 10 runs, while
the C++ version (using VC++ 5 and the SGI STL) averaged just over 3 seconds.
Recently I've run the Java version of the benchmark again, using both
today's standard VM (Sun 1.2.2 with Symantec JIT), and some more exotic Java
compilation technology (NaturalBridge BulletTrain). The results were
interesting. The 1.2.2 JIT was a bit of a disappointment, but the
BulletTrain result was, I thought, remarkable. Here are the times (which
didn't vary much over several runs):
JDK 1.2.2: 12819 milliseconds
BulletTrain: 2814 milliseconds
Lest anyone think we have a new champion, I should mention that Allegro
Common Lisp 5.01 Lite is clocking in at around 2.7 seconds, and I haven't
done anything with C++ yet this time around. OTOH I haven't yet tested JOVE
or HotSpot either
</forwarded>
"F. Xavier Noria" <···@retemail.es> wrote in message
·······································@4ax.com...
> The post I forward (from a java-related newsgroup) praises somehow
> the speed of Allegro Lite.
>
> From: "Bruce Tobin" <······@columbus.rr.com>
>
a
> benchmark based on the CN2 rule induction algorithm.
> JDK 1.2.2: 12819 milliseconds
>
> BulletTrain: 2814 milliseconds
>
> Lest anyone think we have a new champion, I should mention that Allegro
> Common Lisp 5.01 Lite is clocking in at around 2.7 seconds, and I haven't
> done anything with C++ yet this time around.
Well, now I have. VC++ (now at 6) with the HP STL is as slow as ever,
averaging around 4 seconds. I suppose I'll have to try with the SGI STL
again, but I don't expect to beat three seconds.
> OTOH I haven't yet tested JOVE
> or HotSpot either
HotSpot, it turns out, is very, very fast. Average under 2 seconds.
"Bruce Tobin" <······@columbus.rr.com> writes:
> "F. Xavier Noria" <···@retemail.es> wrote in message
> ·······································@4ax.com...
> > The post I forward (from a java-related newsgroup) praises somehow
> > the speed of Allegro Lite.
> >
> > From: "Bruce Tobin" <······@columbus.rr.com>
> >
> a
> > benchmark based on the CN2 rule induction algorithm.
>
>
> > JDK 1.2.2: 12819 milliseconds
> >
> > BulletTrain: 2814 milliseconds
> >
> > Lest anyone think we have a new champion, I should mention that Allegro
> > Common Lisp 5.01 Lite is clocking in at around 2.7 seconds, and I haven't
> > done anything with C++ yet this time around.
>
> Well, now I have. VC++ (now at 6) with the HP STL is as slow as ever,
> averaging around 4 seconds. I suppose I'll have to try with the SGI STL
> again, but I don't expect to beat three seconds.
>
> > OTOH I haven't yet tested JOVE
> > or HotSpot either
>
> HotSpot, it turns out, is very, very fast. Average under 2 seconds.
>
Has anybody tried CMUCL of this test?
Cheers
--
Marco Antoniotti ===========================================
PARADES, Via San Pantaleo 66, I-00186 Rome, ITALY
tel. +39 - 06 68 10 03 17, fax. +39 - 06 68 80 79 26
http://www.parades.rm.cnr.it/~marcoxa