[1378] in Coldmud discussion meeting

root meeting help first first in chain previous in chain previous next next in chain last in chain last

Re: [COLD] NBC Online project

daemon@ATHENA.MIT.EDU (Thu Nov 20 12:09:06 1997 )

From: Jay Carlson <nop@nop.com>
To: Frank Crowell <frankc@maddog.com>
cc: Cold Stuff <coldstuff@cold.org>
In-reply-to: Your message of "Wed, 19 Nov 1997 23:56:04 -0900."
             <3473FB24.CABF9998@maddog.com> 
Date: Thu, 20 Nov 1997 12:01:38 -0500

> There are also some technical problems.  They want the mud to be able to
> handle 300+ online users. I believe that LambdaMOO lags to an impossible
> level at over 200-- 

Depends on what you're counting.  LambdaMOO regularly hits 200 users
connected with quite acceptable lag, but a good number of those 200
users are idling.  

Looking through the last week or so of stats, I found the high water
mark for non-idle players was 151, with average lag at that point of
about 1.7s.  The server's been up for over three weeks now, so we're
seeing a little bit of a performance hit from various kinds of
fragmentation.

(LambdaMOO tends to overflow my verbcall counter about three weeks
into a run.  I should probably rewrite the report generator but it
amuses me to see -1,949,136,190 hits on the verb cache....)

Anyway, lambda.moo.mud.org doesn't seem like a good measure of what
the bounds on MOO performance are.  It runs on a 4-CPU SS1000; that
was a pretty spiffy machine a few years ago but today's el-cheapo
Cyrix P166 (like the one I have at home) has better pure integer
performance.  What's more, lambda.moo.mud.org doesn't have enough
memory---256M of core for a rather active 370M process is just asking
for bursty lag.  No, I don't think it would be prudent to gloat about
how disk-basing wouldn't have this problem until there's a concrete
example in the same ballpark....

Every now and then I spec out a "replacement" machine for LambdaMOO.
Depending on vendor reputation, I get between $6-11k.  I suspect
performance would easily triple on a well-built multiprocessor PPro 200.
-- 
Jay Carlson    nop@nop.com    nop@mitre.org

Flat text is just *never* what you want.   ---stephen p spackman