[Coldstuff] Re: Bytecoded language speed

[coldstuff@cold.org]

On Wed, 17 July 2002, "Frank Crowell" wrote

> I am not clear on what it would take to handle a heavily populated, animated
> world.  I saw the death of Gods Net (if I remember the name correctly) on
> LambdaMOO die because it was too much for the server.  Even LPMuds started
> to show signs of poor performance when they had mobs that did too much.
> Dikus probably did a little better because there was no interpeted coded.
> 
> At some point maybe this becames a server issue and not a database issue.

Much of the problem here is that bytecode MUDs are very easy to code on, and
they're very fast before people start logging on.  Everyone gets so wrapped up
with writing cool code quickly, nobody pays attention to speed.  However, I do
think that bytecode MUDs can run efficiently if speed is kept in mind from the
start.

ColdC incorporates data types and programming structures that would be a major
bi*ch to implement in languages such as C, making C programmers much less
likely to use these optimized methods.  With things like dictionaries being so
easy to implement in ColdC, a speed-conscious programmer can easily make code
that is actually faster because it is easier to "do it right".

Speed is a major concern on Epoch, which uses dictionaries to find commands. 
I don't know how this compares to other cores, but it takes about 6K ticks to
find a command in a populated room, and 4K to execute it (including parsing
markup) on average.  That's about 55K microseconds on my Pent/133.  Yes,
statistics are automatically kept on this.

Anyone have any other benchmarks?

- Kipp