Thread: [tcljava-user] Tclblend performance

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 422-6466

Is running java through tclblend slower that pure java, and if so, how much?

Klara Ward

On Thu, 3 Jan 2002, Klara Ward wrote:
> Is running java through tclblend slower that pure java, and if so, how much?

The tclblend <-> java connection is done through reflection and JNI and
thus has the overhead of those API's.  I've heard figures like a method
call through reflection is 10-20x slower than a normal Java method call.

The way to work around this problem, then, is to reduce the number of
calls across the Tcl/Java interface -- in other words call a fairly
coarse-grained Java method that does a lot of work rather than making lots
of little method calls.

Of course, this is easily measurable -- just write a Tcl script that times
a couple hundred thousand method calls to a Java object, then write a
little Java program that does the same, and compare the times.

			-=- D. J.

So that would mean that only method calls from tcl to java are slow?
I changed a program from invoking a bat-file running tcl-code from java to
creating a java object from the tcl code and it seems much slower, tcl and
java then communicates with a socket in both cases(java invoke tcl, tcl
invoke java).

Klara

----- Original Message -----
From: "D. J. Hagberg" <dha...@mi...>
To: <tcl...@li...>
Sent: Thursday, January 03, 2002 5:06 PM
Subject: Re: [tcljava-user] Tclblend performance

> On Thu, 3 Jan 2002, Klara Ward wrote:
> > Is running java through tclblend slower that pure java, and if so, how
much?
>
> The tclblend <-> java connection is done through reflection and JNI and
> thus has the overhead of those API's.  I've heard figures like a method
> call through reflection is 10-20x slower than a normal Java method call.
>
> The way to work around this problem, then, is to reduce the number of
> calls across the Tcl/Java interface -- in other words call a fairly
> coarse-grained Java method that does a lot of work rather than making lots
> of little method calls.
>
> Of course, this is easily measurable -- just write a Tcl script that times
> a couple hundred thousand method calls to a Java object, then write a
> little Java program that does the same, and compare the times.
>
> -=- D. J.
>
>
>
> _______________________________________________
> tcljava-user mailing list
> tcl...@li...
> https://lists.sourceforge.net/lists/listinfo/tcljava-user
>

On Fri, 4 Jan 2002 16:23:38 +0100
"Klara Ward" <kl...@sp...> wrote:

> So that would mean that only method calls from tcl to java are slow?

As far as I knew, the topic was comparing invocation of a Java method
from Tcl via the JNI to invoking a Java method directly from Java code.
Of course the latter is going to be slower, there is a lot more the system
has to do. Comparing a Java method invocation to a call to exec is
apples and oranges.

> I changed a program from invoking a bat-file running tcl-code from java to
> creating a java object from the tcl code and it seems much slower, tcl and
> java then communicates with a socket in both cases(java invoke tcl, tcl
> invoke java).

It sounds like you were execing a Tcl process from inside Java via Runtime.exec().
There is no way that would be faster than calling a Java method on an existing
Java object. Your code does not use sockets after you switched to Tcl Blend, right?
Perhaps if you trim what you are trying to do down to a small/simple example and
post the code the details will be more clear. You could also try some simple timing
numbers using the time command in Tcl.

Mo