|
From: | Andy Cleary |
Subject: | Re: [Swarm-Modelling] SWARM on Clusters |
Date: | Thu, 29 Jan 2004 13:26:42 -0800 |
At 01:57 PM 1/29/2004 -0700, you wrote:
address@hidden wrote:The cost comes at event insert time. For most folks this is when the program plans what it wants to do (e.g. buildActions) or if the simulation is doing dynamic scheduling. In normal situations, the figure-out-what-to-do-next cost is neglible.When it comes down to executing, Swarm has to go through all the events and figure out which one runs next.BTW clusters don't help much. They are designed to distribute the threads of aEven if Swarm was adapted to be multithreaded, a cluster wouldn't help due to communication costs. Linux extensions like MOSIX work fine for migrating simple processes from machine to machine, but if the process that wants to be migrated has significant memory resources or I/O activity it won't move. So, for example, agents that shared a landscape that was in RAM (especially swapping on RAM) on one computer would often get stuck on the computer having the memory resource, even though there were thousands of agents that would like to have a processor to themselves.multi-threaded application. For all intents and purposes, Swarm is a single threaded application (when it comes to executing the model). So a cluster doesn't help at all.
You just can't do any reasonable level of parallelism without taking a more explicit "partitioned/domain decomposed" approach, and for that a new generation of ABM infrastructure is needed...
Andy
_______________________________________________ Modelling mailing list address@hidden http://www.swarm.org/mailman/listinfo/modelling
[Prev in Thread] | Current Thread | [Next in Thread] |