Parallel Question and Swarm Future

[donalson]

This is a 2-part question/statement.  The first is about the mythical
parallel version of Swarm and how it could help model speed, and the
second is about the future of Swarm.

As I understand parallelization (sp? Or even word?) as it pertains to a
Swarm type simulation, is distributed computing.  That is, the
simulation is broken down into subsections and “farmed out” to different
processors and/or computers.  There is a gain in total simulation speed
when the savings from running bits of the sim in parallel exceeds the
cost of the communication between processors.  The problem I have here
is the nature of (at least my) simulations.  For example, a reasonable
way to distribute the sim is by breaking up space into subsections and
farming them out.  Now, the obvious reason to have space in the system
is because it is important and this usually means movement is important
as well.  This implies that some sort of granularity must be imposed on
the system for it to be broken up i.e. Each subprocess execute for 10
time steps and then re-distribute the agents.  The larger the “grain” or
step size the better the speedup.  The problem with increasing the step
size is that it begins to impose a false time scale on the dynamics.
The better the speed up, the worse the problem.  Now, my sims take as
long as 3 weeks to 1 month to run, so I am very anxious for any
speedup.  Is there something false in my logic?  Am I missing
something?  This is one case where I would love to be wrong…

WARNING:  I made this second part controversial on purpose, it is meant
to get a discussion going.

Second, and I will bring this up because Marcus mentioned it in passing
as well about a month ago, is the question of the Swarm collections
library.  I say this with all deference to Roger Burkhart and the other
early developers who put an amazing amount of time into the project and,
I suspect, never thought there work would be so good as to go this far.
I owe them a lot, ½ of my dissertation, and a lot of my rep is based on
their work.

Given that, the collections library needs an overhaul.  This is a
significant problem as all of the scheduling code is based on
collections.  It schedule routines are also very complex and rather
fragile.  It seems to me that if Swarm is going to continue, this is the
next big step.  As Jan remarked, we need a better faster engine
(parallel?) as opposed to more graphing capabilities.  If I remember
correctly, I was introduced to Manor when he first joined the Swarm team
as the person who was going to parallalize Swarm.  How long ago was
that?  I am taking a wild guess that this has still not been
accomplished because, as things stand, it is too difficult.

The question is then, do we continue to build on an unstable foundation,
or do we build for the future.  In addition, for all the good reasons
Swarm was built in Objective C, it is still serious baggage because of
the lack of development tools and compiler support available.  Prior to
Swarm, I was using Centerline’s C/C++ compiler and debugger.  I still
get tears in my eyes when I think about having had to give it up.  Is it
too much to think about a translation to C++?  What are the consequences
of keeping ObjC?  This is not a idle question from my standpoint, as I
have been hired to build spatial models of the intertidal zone for Cal
State LA and I have to make decisions on the machines we are going to
purchase and the language I will write in.  People become concerned when
you write in a language that is little used.  Code upgradability and
maintainability become important in long term projects and very few
people are fluent on ObjC.

All right folks, flame away…

D3


--
*********************************************************************
* Doug Donalson                 Office: (805) 893-2962
* Ecology, Evolution,           Home:   (805) 961-4447
* and Marine Biology            email address@hidden
* UC Santa Barbara
* Santa Barbara Ca. 93106
*********************************************************************
*
*   The most exciting phrase to hear in science, the one that
*   heralds new discoveries, is not "EUREKA" (I have found it) but
*   "That's funny ...?"
*
*       Isaac Asimov
*
*********************************************************************




                  ==================================
   Swarm-Modelling is for discussion of Simulation and Modelling techniques
   esp. using Swarm.  For list administration needs (esp. [un]subscribing),
   please send a message to <address@hidden> with "help" in the
   body of the message.
                  ==================================