Menu
▾
▴
RE: [Repast-developer] Controller hierarchy
|
From: Laszlo G. <gu...@la...> - 2002-12-13 17:39:28
|
Hi, Thanks for the clarifying notes. While I think both goals seem worth pursuing, I was more like thinking about item #1. Now that I come to think about it, I have a CS undergraduate I advise back in Hungary, who already put a simple, but modular, Condor-like framework together in Java. I might redirect him to hook it up with Repast. This is in addition to another CS undergrad. who has agreed to work on 'something for Repast'. Maybe, the two of them could round up nicely what Tom has in mind.... I'll talk to them in January and get back to you! Gulya At 10:27 AM 12/13/2002 -0500, Nick Collier wrote: >I also want to say that we have two goals with respect to distributed >simulation. > >1. Do multiple batch runs on multiple computers from a single computer, >collecting the data on that single computer. The idea here is to split >the parameter space into pieces and explore these spaces individually on >individual computers. This can be started and controlled from a single >computer and the results should be recorded on that computer. Of course, >something like this can already be done with Drone or Drone-like >toolkits. However, our target hardware here is something like a >computing lab full of win2K machines that are otherwise idle over night, >or 4 or 5 scrounged and freely available windows boxes. > >This is not that difficult and shouldn't require a parallel distributed >scheduler. > >2. Spread a single simulation run over multiple computers. This is more >traditional distributed parallel computing and would require a >distributed parallel scheduler. > >I'm not sure where we've decided to place our focus w/r to these two >goals, although I know my heart and my head are divided here! > >Nick > >On Fri, 2002-12-13 at 10:09, Nick Collier wrote: > > Hi all, > > > > I think there's a bit of confusion here. What's included in repast 2.0 > > is not Mike's FAST framework (a framework for distributed repast -- more > > or less -- simulations) but rather code that enables the true > > parallelization of scheduled actions. The intention is that expensive or > > long running actions can be scheduled to run in parallel with typical > > sequential scheduled actions. > > > > Maybe Mike can say more about the relationship of something like FAST to > > these new parallel actions. > > > > Anyway, my point is that repast 2.0 doesn't give you framework for fully > > distributed parallel simulations out of the box. We are pursuing > > something like this, however. Tom has looked into Globus, Condor and > > lately ProActive (http://www-sop.inria.fr/oasis/ProActive/) in this > > context. I think it will be a priority for him once he gets the current > > round of GIS code sorted. Maybe he can say more, although I know he's > > away for a few days at the moment. > > > > Nick > > > > On Fri, 2002-12-13 at 09:55, Frank Blecha wrote: > > > > > > I'm also curious as to the inclusion of that feature. - Frank > > > > > > On Thu, 12 Dec 2002, Scholand, Andrew J wrote: > > > > > > > Hi, > > > > > > > > Work in this vein is often termed 'embarrassingly parallel' or > parametric > > > > computing - see for example NimRod/G, > > > > http://www.globus.org/research/applications/nimrod.html. Might be > a good > > > > tie in with Mike North's parallelized RePast engines, which I > believe will > > > > be in 2.0? > > > > > > > > Cheers, > > > > Andy > > > > > > > > -----Original Message----- > > > > From: Skye Bender-deMoll > > > > To: rep...@li...; nic...@ve... > > > > Sent: 12/12/2002 1:11 PM > > > > Subject: Re: [Repast-developer] Controller hierarchy > > > > > > > > > > > > Hi nick, and folks, > > > > Was just remembering a conversation we had a long time ago about > > > > the idea of a ParameterSpaceExplorer, which seems relevant to the > > > > controller discussion. > > > > > > > > Crudely, the idea is pretty much the same as doing runs from a batch > > > > file, but the batch file has a GUI and data presentation of its > > > > results. So after someone has finished coding a model, and wants to > > > > explore "results" over a particular range of parameters, they could > > > > load the ParameterSpaceExplorer, enter the range, parameter sampling > > > > density, and number of repetitions per point, and come back in a > > > > couple of hours to a nicely collected data set. Of course this > > > > requires that models have well defined stopping conditions (or a set > > > > number of steps), and report clear numeric results of some kinds. > > > > > > > > In the (rare) case where there are two main input parameters of > > > > interest, and one result parameter, we could have nice 2.5D plots of > > > > results, or if multiple result params, allow switching between result > > > > plots... Could also calc "pseduo"-error ranges and stdv. Then > > > > the user can say, "hmm, what's this discontinuity on the graph, is it > > > > just randomness? lets do another 300 runs, more closely spaced along > > > > parameter-x, and see what we get..click click.." > > > > > > > > Another advantage is that it might help people (us model coders) > > > > focus their (our) thinking on "OK, what are my inputs, and what are > > > > my results, and what does it mean" instead of just, "WOW! isn't that > > > > pretty!" I don't by any means mean to dis good graphical > > > > representations of a sim, as I think they are incredibly important, > > > > but I think it is also good to help formulate a "question" and > > > > "results" approach as well. > > > > > > > > I know it is better to do things than suggest them, but I've got all > > > > the java I can handle at the moment with this network stuff. But I > > > > thought I'd throw the idea out there again, just in case coordination > > > > between DataSource, DataRecorder, and the Controller interface might > > > > be useful or necessary in the future... > > > > > > > > cheers, > > > > -skye > > > > > > > > > > > > > > > > > > > > ------------------------------------------------------- > > > > This sf.net email is sponsored by: > > > > With Great Power, Comes Great Responsibility > > > > Learn to use your power at OSDN's High Performance Computing Channel > > > > http://hpc.devchannel.org/ > > > > _______________________________________________ > > > > Repast-developer mailing list > > > > Rep...@li... > > > > https://lists.sourceforge.net/lists/listinfo/repast-developer > > > > > > > > > -- > > Nick Collier > > Social Science Research Computing > > University of Chicago > > http://repast.sourceforge.net > > > > > > > > ------------------------------------------------------- > > This sf.net email is sponsored by: > > With Great Power, Comes Great Responsibility > > Learn to use your power at OSDN's High Performance Computing Channel > > http://hpc.devchannel.org/ > > _______________________________________________ > > Repast-developer mailing list > > Rep...@li... > > https://lists.sourceforge.net/lists/listinfo/repast-developer >-- >Nick Collier >Social Science Research Computing >University of Chicago >http://repast.sourceforge.net > > > >------------------------------------------------------- >This sf.net email is sponsored by: >With Great Power, Comes Great Responsibility >Learn to use your power at OSDN's High Performance Computing Channel >http://hpc.devchannel.org/ >_______________________________________________ >Repast-developer mailing list >Rep...@li... >https://lists.sourceforge.net/lists/listinfo/repast-developer |
