Well, I have downloaded my first work unit for ACH (AfricanClimate@Home) so I should start running it on my Windows XP SP2 computer in about 3 hours. I suppose I might as well write a post explaining what I think this project is about, so people can correct my mistakes.

Reading the Member News posted by knreed ( http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=16044 ), the project description ( http://www.worldcommunitygrid.org/projects_showcase/ach/viewAchAbout.do and http://www.worldcommunitygrid.org/projects_sh...ewAchFaq.do?shortName=ach and http://www.worldcommunitygrid.org/projects_showcase/ach/viewAchPart.do ), the project website ( http://www.csag.uct.ac.za/ and http://www.csag.uct.ac.za/worldcommunitygrid ) and the WRF (Weather Research and Forecasting Model) site at http://www.wrf-model.org/index.php , I have some ideas that I am hazy about.

ACH is trying to model the climate around South Africa. This is a Regional Climate Model. Reading the WRF site, the programs split up the world into tiles. Are our work units treating all of Southern Africa as one tile, meaning that each work unit spans the entire region, or is each work unit assigned a much smaller area within the Southern Africa region?

I have heard that each work unit will model 2 weeks of weather and that each trial model will be iterated 27 times, covering a span of 378 days. This brings up some vocabulary questions, A model will be broken up into work units, then run to span 2 weeks. After all the results are back, they will be used to create new work units to span the next 2 weeks and those will be sent out. Right now, I am calling that an iteration. What should I be calling it? For that matter, I am calling the overall run a 'model'. Is that right?

I assume that we will be running many different models. The most obvious possibilities are past years with known weather, to check the model calculations and allow improvements to be made. But just what other sorts of models might we be running?

And I have a totally off-the-wall question. How much land does our model cover and how much ocean is included? This does not make any difference to me personally. It is just an example of the sort of questions that pop into my mind.

Another question is if there is any possibility that we will eventually make multi-year runs? Or do the current applications produce results that vary farther and farther from reality over longer time spans?

As I interpret the post in Member News, the final result is about 130 MB in size. This is a monster upload, especially since most Internet connections upload much slower than they download. Apparently the entire result file is hashed (a small number is produced by running a mathematical function over the entire result file) and each copy of the work unit sends back the hash number of its result file. If all the computers assigned the work unit send back the same hash number, then they all produced identical result files. The clever idea in this project is to have each computer assigned a specific tenth of the result file to send back. So 10 computers each receive the same work unit, but only send back 13 MB instead of about 130 MB. This is where my questions start. If one computer produces a different hash number and it was assigned (for example) the 5th part of the result to send back (out of 10 total), does the server resend the work unit to an 11th computer with an order to return the 5th part of the result? I am sure that knreed has been pondering all the different possibilities about just when to send out makeup copies and how many. Experience might lead to some changes, but it would be interesting to hear what we are dong at the very start.

Just thinking about the communication problems with all these uploads and downloads makes me glad that this is an opt-in project that nobody will get unless they decide to check it in 'My Projects'. It is definitely not for everybody.

But I am about to give it a try!


Lawrence