MOO-cows Mailing List Archive


Re: chat room

At 01:47 PM 4/11/95 PST, Microwave Systems Admin wrote:
>On Thu, 2 Nov 1995, Tom Ritchford wrote:
>> From: (Jeanne Davis)
>> >Hi, have been lurking on the list for awhile and have yet to see this
>> >question answered.  How many people can be present in a room in a moo based
>> >on the LambdaCore at once?  My boss wants the moo I am working on to have
>> >an arena (read AOL) style chat room, which can accomodate 1000s of users.
>> >Could they all be logged into the same room at the same time without
>> >compromising the stability of the moo?
>> Interesting idea... I love testing the limits on things.
>> AFAIK there is no theoretical limit on the number of people in a single
>> room.  However, one would think that the load on the machine would go
>> up as the *square* of the number of users (since each player in the room
>> could be generating text and that would have to be sent to each other
>> player...)  That would mean that 1000 users in a room would use on the
>> approximate order of 1,000,000 as many tics as a single user...  you'd
>> need a VERY fast machine indeed!
>Actually, if one used a very efficient :notify and/or :tell verb, a 
>normally available unix system would probably be able to handle this easily..

We crashed our Sun Sparcstation20 early on in our MOO experience by
having 30 people log in at once.

 - at that time there was some sort of memory-wasting going on and
   our MOO was taking up about 15Mb of RAM while running.
   It now is more like 5Mb.

- the crash left the moo running, but all Network interfaces (except loopback)
  on the Sun w/- Solaris 2.4 were dead.  I had to drive out to the
  installation and reboot to get things running again.  Before I rebooted I
  connecting to the MOO from the console and it was still running, but the
  box's 4 external network interfaces were all dead.

- we had invited 30 people sitting in a room to telnet to the MOO
  It seemed to crash after about 25 or so.  We've since had more users than
  that connected without problems, so this isn't an accurate user limit.

I'd guess our crash was aggrevated by the memory-hogging behaviour of the MOO
at the time, and perhaps by the fact that 30 ppl were all connecting
We haven't had an opportunity to try the same thing again, esp. since there
are other users of the box who don't appreciate us crashing it.

If I was thinking about a 1000-user moo room I would want to be sure
the server could handle 1000 connections - seems to me you might need
to create some extra Pseudo TTY devices.

As far a O(n^2) performance goes, the good classroom's I've seen use "tables"
to limit the propogation of most chat, so you essentially have a lot of little
rooms (the tables) for ~10 people, combined into a big room.  Only the speaker
broadcasts to everyone.  Questions are queued and then broadcast to everyone
as they are answered.  This should give excellent scalability.


Guy Carpenter - Clearwater Technical Services
Tel: (070) 953309                  Fax: (070) 953007

Home | Subject Index | Thread Index