[MUD-Dev] Re: [Java] multithreading: update and a question

[Ben Greear]

On Wed, 15 Jul 1998, Vadim Tkachenko wrote:

> Chris Gray wrote:
> 
> [skipped]
> 
> > I'd guess that whether or not you want one OS thread per Java thread
> > would end up depending on how many Java threads you use, and how often
> > you create and destroy them.
> 
> [speaking about me]
> 
> I'm creating and destroying threads deliberately carelessly, on demand,
> 'cause remember that the preliminary optimization is a mother of all
> evil. The Jukebox as it is today survived almost 2 years before I
> started (because needed to) to optimize it.
> 
> > If you end up with only about a dozen or less,
> 
> Nah, the estimate is a count of hundreds - based on the test results, I
> consider a 1000 threads a reasonable default for a number of threads -
> that was Linux & Solaris on PII/350 with 128M RAM, which is I guess not
> considered a server configuration novadays.

As much as I love Linux, at the present time it is not ready for running
CPU intensive (or maybe any other kind of intensive) java programs.  I get
markedly better performance under Win95, and Solaris, which is to be
expected, considering JIT etc...

Hopefully, this will change soon.  Corel said they were gonna work on
a JIT for Linux (for their Netwinder).

> See, there's a drawback - the architecture solutions for one/one,
> many/one and many/many are different. Java becomes a curse
> portability-wise, because it's SUPPOSED to behave in the same way, but
> depending on the platform implementation different people working on
> different platforms will lose their (voice fibers? what's the English
> term? am I making myself clear here?) shouting and trying to convince
> each other in what sucks and what rules.

Vocal cords I think you meant! :)

> -- 
> Still alive and smile stays on,
> Vadim Tkachenko <vt at freehold.crocodile.org>

After seeing the numbers for Clustering (ie the Beowulf Supercomputer
they built at NASA), I think we might see clustered boxes before
wide-spread dual+ CPU stuff.  This will probably remain in the Linux
world, as opposed to Win95 (NT might cluster one day.)

Hrm, "thread" at process level, use UDP or TCP for communication layer,
seems very portable to me.  Can always use lighter threads within each
process thread..  Then can take advantage of clustering and or multi-cpu
boxes...

Hrm, not making much sense even to myself....

Ben


Ben Greear (greear at cyberhighway.net)  http://www.primenet.com/~greear 
Author of ScryMUD:  mud.primenet.com 4444
http://www.primenet.com/~greear/ScryMUD/scry.html