swarm-support
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: Another interesting issue ....


From: Steven Clark
Subject: RE: Another interesting issue ....
Date: Tue, 27 May 1997 16:43:17 -0400

There are a lot of issues Swarm has addressed (successfully) that have
shaped its design.  Your (Mr. Jojic's) suggestion of making Swarm into a
distributed system is very interesting to me, although my feeling is
that it would be necessary to re-think some basic design elements of
Swarm.

Prior to learning Swarm, I did some agent programming on a system that
may now be commercially available.  It is called Cybele, reference
http://www.i-a-i.com//projects/cybele/index.html
(The web server is on a slow connection to the Net and the pages appear
to have some bugs; I had trouble but finally got it.)

I believe for a distributed system as you and I envision it - with 100's
or 1000's of agents per computer - to achieve good performance it is
necessary to do a handful of interesting things:
1.  Get away from the RPC model and use Email as your messaging model.
The message sender does not block when it sends a message.  Instead, if
necessary, the receiver sends a message back.
2.  Use broadcast or multicast message sending when appropriate.  (This
is the newsgroup model for message passing.)  Implement it with true
broadcast/multicast at the network level (something rarely done).
3.  Make message addressing independent of destination agent's location.
4.  Find a distributed way to do what Swarm does in its centralized
schedule.

Cybele does 1-3.  It is possible to circumvent 4 for certain problems.
The version of Cybele that I used was coincidentally Objective-C based,
although it ran on NeXT rather than Gnu.

There is a fairly large amount of work involved in getting it all
together.  For example, a computer needs to know if it should put a
message out on the net or if its recipient(s) is(are) local, and yet
agents can be created, destroyed, and possibly migrated.

Personally, I don't see a wedding of these two approaches (Swarm and
Cybele) in the near future.  They are oriented towards different problem
sets.  When we can afford to waste more computing power, we will be able
to afford a common solution.

Steven J. Clark       address@hidden       313-769-4396
Center for Electronic Commerce, Industrial Technology Institute
Box 1485, Ann Arbor, MI  48106

[[Mr. Jojik's posting about making Swarm into a distributed system]]

                  ==================================
   Swarm-Support is for discussion of the technical details of the day
   to day usage of Swarm.  For list administration needs (esp.
   [un]subscribing), please send a message to <address@hidden>
   with "help" in the body of the message.
                  ==================================


reply via email to

[Prev in Thread] Current Thread [Next in Thread]