Re-4: [ObjectWeb architecture] just an idea for (Java) native compilation

[<markus.karg@xxxxxxxxx>]

Thank you for the links. Actually what that guy did was micro benchmarking. 
This is unfair for java and other late-compiling VMs: What that guy measures 
includes the startuptime of the VM, also measures interpreted (!) code (not 
hot spot compiled code). To measure hot spot compiled code, you need to put 
all functions into one, execute that function one million times, then look 
the average value. That's a real life test. No one cares on the java.exe 
startup time in a server environment! No one cares the first 100 iterations 
of an algorithm being run in iterpreted mode if the next 24 hours this server 
will execute the same function one million time in hot spot compiled mode.

To write fair measuring micro benchmarks you need to know how hot spot VMs 
work. That guy seems not to know (I told him today). So I'm looking forward 
to see the actual results done with fair tests.

The fact is that Java on a modern hot spot VM (1.4.2, 1.5) with most modern 
GCs (Concurrent GC etc.) are nearly fast as C code. If you don't believe, ask 
Sun's GC team for a fair, real-live test suite.

Microbenchmarks are no proof since no one starts a server's java.exe more 
than once a month, and no one stops the measuring after one run.

This is why Sun and IBM believe in HotSpot. Native compilation is necessary 
only if there would be a real life benchmark failing, not a Microbenchmark. 
Unfortunately, this case did not happen in the past.

Have Fun
Markus

-------- Original Message --------
Subject: RE: Re-2: [ObjectWeb architecture] just an idea for (Java) native 
compilation (25-Sep-2003 15:07)
From:    dominique.devito@xxxxxxxxxxxxxxx
To:      markus.karg@xxxxxxxxx

> >===== Original Message From <markus.karg@xxxxxxxxx> =====
> >Oh, now I understand. So is the OCaml compiled native code faster than the 
> 
> code from other compilers, e. g. from GCJ? Do you have benchmarks (I'm 
> really 
> interested in getting Java running faster)?
> 
> See The Great Computer Language Shootout 
> <http://www.bagley.org/~doug/shootout/> and more precisely, the page 
> <http://www.bagley.org/~doug/shootout/craps.shtml>.
> 
> But I don't know if these benchmarks are what you are expecting.
> 
> I understand what you mean : due to open source solutions, hardware becomes 
> 
> now the more expensive point.
> 
> >For the second point: Sure constructing an
> >ObjectWeb own native java compiler is interesting, but do we have manpower 
> 
> for such a project?
> 
> I am some ideas about it but of course, I don't know it will work :-)
> I will post these ideas in a just-to-come email into the "architecture" 
> mailing list and I will put you in copy.
> 
> Dominique
> 
> 
> >-------- Original Message --------
> >Subject: RE: [ObjectWeb architecture] just an idea for (Java) native 
> compilation (25-Sep-2003 11:08)
> >From:    dominique.devito@xxxxxxxxxxxxxxx
> >To:      markus.karg@xxxxxxxxx
> >
> >> >===== Original Message From <markus.karg@xxxxxxxxx> =====
> >> >Maybe I have not quite understood what the benefit of a cross 
> compilation
> >> from Java to OCaml should be. Java is fast already and if I want to make 
> it
> >>
> >> faster, I would prefer native compilation (machine code) instead of 
> OCaml.
> >> Cross compilation
> >>
> >> I also focus here native compilation for Java but between the Java code 
> and
> >>
> >> the binary code, I have just talked about an intermediary step : the 
> OCaml
> >> code.
> >>
> >>
> >> >between Java and OCaml would weaken the idea of one common Java 
> platform,
> >> since it would degrade Java to one of many possible source languages for
> >> OCaml. But instead, the idea behind Java is that it is the only target
> >> platform itself. So I will not
> >> >support this idea.
> >> >Also OCaml is seldomly used, while Java is widely spread. So I only see 
> a
> >> benefit for OCaml, not for JOnAS.
> >>
> >> No degradation here for my point of view : the OCaml language is used 
> here
> >> as
> >> an intermediary step for Java native compilation. Just as the first C++
> >> compilers. The C language was used for the intermediary step : C++ -> C -
> >
> >> native code and this compilation way has not weakened the C++ language 
> and
> >> it
> >> was not for the benefits of C.
> >>
> >> Instead of generating OCaml code from the Java code, another way might 
> be
> >> to
> >> reuse&share some parts of the outstanding OCaml compiler in order to 
> build
> >> a
> >> Java compiler.
> >>
> >> Just my 2 cents.
> >>
> >> Best regards,
> >> Dominique De Vito
> >>
> >>
> >> >Have Fun
> >> >Markus
> >> >
> >> >-------- Original Message --------
> >> >Subject: [ObjectWeb architecture] just an idea for (Java) native
> >> compilation
> >> (24-Sep-2003 21:56)
> >> >From:    dominique.devito@xxxxxxxxxxxxxxx
> >> >To:      markus.karg@xxxxxxxxx
> >> >
> >> >>
> >> >> Hi,
> >> >>
> >> >> This is just an idea for (Java) native compilation.
> >> >>
> >> >> INRIA has an outstanding knowledge about native compiler technology. 
> To
> >> be
> >> >> more precise, INRIA has developped a native compiler for the "
> Objective
> >> >> Caml" language. Caml is a strongly-typed functional programming 
> language
> >> >> from the ML family. OCaml (Objective Caml) and Caml Light are two 
> open
> >> >> source implementations of Caml developed at INRIA Rocquencourt, 
> projet
> >> >> Cristal. See <http://caml.inria.fr/> and <http://caml.inria.fr/ocaml/>
> .
> >> >>
> >> >> INRIA has started at January 2001 the Caml consortium (like the
> >> ObjectWeb
> >> >> consortium) : the Caml consortium federates the design and 
> development
> >> of
> >> >> the Caml language and its programming environment. The Objective Caml 
> (
> >> >> OCaml) compiler is released by the Caml Consortium and performs very
> >> well :
> >> >> Objective Caml ranks 2nd on speed (between C and C++) on Doug Bagley'
> s
> >> >> computer language shootout ! See 
> <http://www.bagley.org/~doug/shootout/>.
> >>
> >> >>
> >> >> So, why not joining Caml & ObjectWeb consortiums strengths for
> >> developping
> >> >> a Java compiler ? One way could/would be to develop a bridge Java->
> OCaml
> >> as
> >> >> OCaml's object-oriented features are similar to Java's object-
> oriented
> >> >> features (Ocaml also includes functional features, module features...)
> .
> >> >>
> >> >> What do you think about this idea ? Silly idea :-) ? Good idea ?
> >> >>
> >> >> Best regards,
> >> >> Dominique
> >> >>
> >> >>
> >> >>
> >> >> To: architecture@xxxxxxxxxxxxx
> >>
> >>
> >>
> >>
> >> To: markus.karg@xxxxxxxxx
> >>     architecture@xxxxxxxxxxxxx
> 
> 
> To: markus.karg@xxxxxxxxx