Page 1 of 1

Some Q & A about the GRID with a expert

Posted: Thu Dec 17, 2009 12:19 pm
by Xymox
[04:11] <Xymox> So maswan, are the experiments at LHC still doing a lot of event analyzing ? or are they done ? can you tell ?
[04:11] <Xymox> Just wondering if they are finished analyzing all the event data yet
[04:12] <maswan> Atlas is going to do a reconstruction of all the data taken starting on the 22nd
[04:13] <chriwi> does that mean reevaluation with new rules?
[04:13] <maswan> It means going over the raw data and creating new derived data with improved parameters and software to better pick up on what happened
[04:14] <chriwi> yes, thats what I mean
[04:14] <Xymox> interesting..
[04:14] <maswan> I think that actual end-user analysis have been pretty sparse
[04:14] <chriwi> it is interesting to watch that video with the woman from Atlas about this topic
[04:15] <Xymox> which vid ?
[04:15] <chriwi> what does end user analysis mean
[04:15] <maswan> It is what physicists do when looking for interesting stuff.
[04:15] <maswan> It's called analysis when they analyze the derived data and try to see something neat
[04:16] <chriwi> ... l=de&emb=1, here you go
[04:17] <Xymox> wow well you can see a interesting perspective on this. You can see the flow of analysis..
[04:20] <Xymox> maswan, I saw a lot of activity on the GRID yesterday. Who uses it the most ? what things are most popular ?
[04:21] <maswan> Hm. Hard to tell for the grid as a whole, I mostly only see what's going on in my corner of it. But there it is still lots and lots of monte carlo simulations that they will then compare to the actual data
[04:22] <Xymox> Do you know what the state of completion of the whole GRID is ? I know CERN had some number of CPU's as a goal. Have they reached it yet ?
[04:23] <maswan> Roughly, yes. But then this expends for each year too.
[04:24] <Xymox> so they are completely ready for physics from the experiment then. Good !..
[04:24] <robcrazee> what hardware are they running on the grid then?
[04:25] <chriwi> ... 9570342668
[04:25] <maswan> robcrazee: normal intel and amd servers
[04:25] <Xymox> yes thats a FUN tour..
[04:25] <maswan> Xymox: Well, not all of the bits are perfectly ready. It was shown this spring that analysis is much more io-intensive than many of the grid clusters are built for.
[04:26] <Xymox> Oh interesting... So upgrade or workaround or something ?
[04:26] <Xymox> Wait what OS ?
[04:26] <maswan> Upgrade, tuning, etc
[04:26] <robcrazee> ok cheers maswan so no unix / aix etc
[04:26] <maswan> Linux, various distros
[04:27] <Xymox> Is the GRID a software ?
[04:28] <chriwi> interesting, that linux already beats all the other unix stuff
[04:28] <maswan> Well, there is multiple grid softwares that have some common protocols.
[04:28] <Xymox> oh right... i knew that.. you explained last time
[04:28] <Xymox> So can any linux app run on the grid ?
[04:29] <maswan> Pretty much, yeah. As long as it is a non-interactive one that just takes some input files and generates some outputs.
[04:29] <Xymox> wow.... thats one poweful set of kit you got there...
[04:30] <Xymox> wait... are there any games installed.. you know, just on the side.. :)
[04:30] <Xymox> say a space wars
[04:31] <maswan> non-interactive games aren't really much fun. :)
[04:31] <chriwi> lol
[04:31] <Xymox> well I assume it could be harnessed where a app could use all those cpus..
[04:31] <chriwi> interaction can only be files or also IP-traffic?
[04:32] <maswan> Some clusters allow that, some don't
[04:32] <maswan> Some only outgoing
[04:32] <Xymox> OoOo so its possible to write a crazy chess playing app
[04:32] <chriwi> I even remember the Game VGA-Planets, the players made the moves on their PC, generated files , sent them to the server and got files back the next day
[04:32] <maswan> Sure
[04:33] <Xymox> hehehehehehe
[04:33] <Xymox> someone should do that
[04:33] <Xymox> GRID chess
[04:33] <chriwi> hehe
[04:33] <Xymox> that would be pretty record setting
[04:34] <chriwi> what about to compile Linix on the grid, would be one CPU for each sourcefile
[04:37] <Xymox> so maybe stupid question.. From a command line, does the GRID work like one big single computer ?
[04:39] <photino> how do you program for grid computation? i mean, resolving conflicts between things carried out at different locations, different latencies etcs."
[04:45] <maswan> Xymox: No, the grid works more like a big batch system, where you define your jobs in terms of what to run and on what input (and what output to save), then submit it to the grid. then wait and eventually you get your answer back