Message boards : Number crunching : What is happening with Geant4?
Message board moderation

To post messages, you must log in.

AuthorMessage
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 11435 - Posted: 22 Nov 2005, 13:32:42 UTC

There has been mention of porting Geant4 to the BOINC platform. Is there any news on the status of this idea?
Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 11435 · Report as offensive     Reply Quote
Desti

Send message
Joined: 16 Jul 05
Posts: 84
Credit: 1,875,851
RAC: 11
Message 11437 - Posted: 22 Nov 2005, 18:08:25 UTC

If I remember correctly, this experiment will start after the LHC is finished and working.
Linux Users Everywhere @ BOINC
[url=http://lhcathome.cern.ch/team_display.php?teamid=717]
ID: 11437 · Report as offensive     Reply Quote
Profile Ben Segal
Volunteer moderator
Project administrator

Send message
Joined: 1 Sep 04
Posts: 119
Credit: 2,579
RAC: 0
Message 11448 - Posted: 23 Nov 2005, 15:16:35 UTC

Hallo MikeW,

Thanks for your interest! A lot of work was done here at CERN on porting GEANT4 to BOINC. The two people who did most of this work are Ignacio Reguero and Juan Antonio Lopez Garcia. They have given me permission to refer you to their documentation on the project, which you'll find at:

https://uimon.cern.ch/twiki/bin/view/LHCAtHome/BoincPort

As you'll see, there's a lot of valuable and detailed information on BOINC porting there.

Best wishes,

Ben Segal
ID: 11448 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 11449 - Posted: 23 Nov 2005, 16:06:16 UTC - in response to Message 11448.  

Hallo MikeW,

Thanks for your interest! A lot of work was done here at CERN on porting GEANT4 to BOINC. The two people who did most of this work are Ignacio Reguero and Juan Antonio Lopez Garcia. They have given me permission to refer you to their documentation on the project, which you'll find at:

https://uimon.cern.ch/twiki/bin/view/LHCAtHome/BoincPort

As you'll see, there's a lot of valuable and detailed information on BOINC porting there.

Best wishes,

Ben Segal


Thanks Ben - it's interesting to see what goes on in CERN. From the documents it seems that your colleagues have largely completed the port. Do you have any idea when, or if, we will see Geant4 work?



Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 11449 · Report as offensive     Reply Quote
Profile Chrulle

Send message
Joined: 27 Jul 04
Posts: 182
Credit: 1,880
RAC: 0
Message 11450 - Posted: 23 Nov 2005, 16:21:43 UTC

Well the software has been ported. The next step is taking one of the models probably the modelling of the Atlas detector and then decoupling the Geant 4 bit from the rest of the Atlas simulation. Figuring out a way to do bits of the simulation on boinc and then finishing it up at cern where the environment is complete. The problem with many of these physics application is that they need on the order of 1 GB of data for a simulation.

Chrulle
Research Assistant & Ex-LHC@home developer
Niels Bohr Institute
ID: 11450 · Report as offensive     Reply Quote
Profile Paul D. Buck

Send message
Joined: 2 Sep 04
Posts: 545
Credit: 148,912
RAC: 0
Message 11452 - Posted: 23 Nov 2005, 22:19:12 UTC - in response to Message 11450.  

Well the software has been ported. The next step is taking one of the models probably the modelling of the Atlas detector and then decoupling the Geant 4 bit from the rest of the Atlas simulation. Figuring out a way to do bits of the simulation on boinc and then finishing it up at cern where the environment is complete. The problem with many of these physics application is that they need on the order of 1 GB of data for a simulation.

Ok, so it takes a little longer to download a work unit ...

When can we get one? :)
ID: 11452 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 11460 - Posted: 24 Nov 2005, 13:45:34 UTC

The real problem with large data files is the bandwidth available from the server.

What bandwidth is needed to support 20,000 clients downloading 1Gb files concurrently? Is a download of 20Tb in five hours actually a sensible use of network capacity?

It seems likely that although the software has been ported, decoupling just this part of the analysis could take quite a lot of effort yet.


Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 11460 · Report as offensive     Reply Quote
Profile Raimund Barbeln
Avatar

Send message
Joined: 13 Jul 05
Posts: 3
Credit: 95,258
RAC: 975
Message 11464 - Posted: 25 Nov 2005, 8:46:56 UTC

Compression of the files would surely help keeping download times and network traffic low.


when life gives you lemons, make lemonade!
ID: 11464 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 11466 - Posted: 25 Nov 2005, 13:10:55 UTC - in response to Message 11464.  
Last modified: 25 Nov 2005, 13:12:14 UTC

Compression of the files would surely help keeping download times and network traffic low.


If you read the documentation Ben Segal refers to you'll find this has already been done. In any case, scientific data often doesn't compress well. A ratio of 2:1 is usually the best one can expect. 20,000 500Mb downloads, or 10Tb, is still a great deal of data to move around.


Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 11466 · Report as offensive     Reply Quote
Michael Karlinsky
Avatar

Send message
Joined: 18 Sep 04
Posts: 163
Credit: 1,192,543
RAC: 0
Message 11467 - Posted: 25 Nov 2005, 13:26:02 UTC - in response to Message 11460.  
Last modified: 25 Nov 2005, 13:27:03 UTC


What bandwidth is needed to support 20,000 clients downloading 1Gb files concurrently? Is a download of 20Tb in five hours actually a sensible use of network capacity?


Just want to make some remarks. (1) Not all 20k clients would be able to
download this large a file, because they are on dialup. (2) If it is possible
to create several WUs from such a file (like Einstein), this would
reduce the load too. (3) Have several UL/DL servers (like Climate && Einstein).

All in all this is feasible.


It seems likely that although the software has been ported, decoupling just this part of the analysis could take quite a lot of effort yet.


They state that they "boincyfied" the application. I may be reading between the lines, but there is no mention of 'checkpointing' and 'progress reporting', just
'init' and 'finish'. So there needs to be done a lot more besides the
decoupling thingy.

Michael
Team Linux Users Everywhere
ID: 11467 · Report as offensive     Reply Quote
Profile adrianxw

Send message
Joined: 29 Sep 04
Posts: 179
Credit: 673,019
RAC: 0
Message 11469 - Posted: 25 Nov 2005, 15:09:16 UTC

>>>
Would the 1 GB of data be used for one WU only, or would it be more like the big data files Einstein sends out?
<<<

This is a key question. If the data is a once only, then using a different delivery technique, something like BitTorrent will ease the strain at the projects servers.

What you could do is seed the BitTorrent and post a link saying it is there rather then changing the BOINC server software. Then people that have the download capabilities could go and grab the Torrent, and those on dial-up would not. You then send a mixture of wu's - the first thing the ATLAS wu's do is see if the Torrent is present, and if not, return immediately with a token credit for the download time.

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

ID: 11469 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 11471 - Posted: 25 Nov 2005, 17:11:51 UTC

Let's not get too carried away.

Chrulle hasn't said that the BOINC results will require 1Gb of data. He said that the physics simulations can require 1Gb.

If the simulation of the ATLAS detector can be split so that the Geant4 simulation can be applied to individual layers or components and the results reassembled before going to the next stage of simulation then substantially less data will need to be transmitted to each host. It's also true that once a machine configuration has been downloaded it needn't be sent again, so only particle data need be sent.

I can see ways that this project can be made to work, but there is a fair amount of work separating the data and reassembling it.

>> there is no mention of 'checkpointing' and 'progress reporting',

This is a major issue. The Lattice project released apps without checkpoints. Some of their results took five to ten days to complete. Leaving a machine working for this long isn't practical for many crunchers and many (myself included) dropped the project as unworkable.


Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 11471 · Report as offensive     Reply Quote
Profile Chrulle

Send message
Joined: 27 Jul 04
Posts: 182
Credit: 1,880
RAC: 0
Message 11474 - Posted: 25 Nov 2005, 20:11:12 UTC

Well, the problem is that when the physicist do a simulation, they compile the model together with the environment which then gives a 1 gb executable. This binary can then be used for a few runs but nothing that would warrant such a large download.

So, Mike is right we have to split the jobs up in a way that the calculation that just need a bit of data is done on BOINC and the bits that really need the entire 1 gb environment are done at CERN.

Or we have to teach the physicist to write modular code and used shared libraries.
The only problem is that that will probably take longer. ;-)

BTW. A few more runs coming along soon.

Chrulle
Research Assistant &amp; Ex-LHC@home developer
Niels Bohr Institute
ID: 11474 · Report as offensive     Reply Quote
Profile adrianxw

Send message
Joined: 29 Sep 04
Posts: 179
Credit: 673,019
RAC: 0
Message 11500 - Posted: 28 Nov 2005, 9:14:22 UTC

>>> Chrulle hasn't said that the BOINC results will require 1Gb of data.

No, but what Ziran and I were trying to introduce to the mix was that if it was necessary to have such large downloads, it may well be possible. Further, that it could be acheived without vast quantities of work as the distribution mechanism already exists, and a simple check in the new client app would be able to detect a run/no run situation straight away.

>>> teach the physicist to write modular code

Yeah, right. I don't think we'll wait for that one!

Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.

ID: 11500 · Report as offensive     Reply Quote
Honza

Send message
Joined: 2 Sep 04
Posts: 7
Credit: 30,712
RAC: 0
Message 11502 - Posted: 28 Nov 2005, 21:29:54 UTC - in response to Message 11474.  

Well, the problem is that when the physicist do a simulation, they compile the model together with the environment which then gives a 1 gb executable. This binary can then be used for a few runs but nothing that would warrant such a large download.

Thanks for clarification.
I was sure about 'which' 1GB we are talking about: WU download (a lot of analyzed adata), WU upload (a lot of results), WU folder (a lot of results generated by application like CPDN), memory usage (still fine with me :-) or perhaps whole slot (application + data, or the enviroment).
So, it is the latest in the list.

A need for split is evident.
Or perhaps, there are some features of BOINC - limit some WU types according to user profile (machine and internet connection specification) or user optional.

I'm sure LHC team will find a way..."The only problem is that that will probably take longer. ;-"
ID: 11502 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 11540 - Posted: 1 Dec 2005, 15:54:27 UTC - in response to Message 11539.  
Last modified: 1 Dec 2005, 15:56:01 UTC

Then the scientists are a bit further along they will probably want to beta test the application. Even if the data files then still are around 1GB, I think that were would be enough people out there that could handle that kind of downloads for a beta test.

"This binary can then be used for a few runs but nothing that would warrant such a large download."

So what is needed is ether a size reduction of the downloaded file or more work done per file. More work per file would mean more time spent on it. How time sensitive would the science be? Then the work/download ratio hits a certain point it probably would be interesting to have a limited production using some of the most productive hosts attached to the project.


I think you've missed the point. The 1Gb executable is a single download for a single purpose, or maybe a few (less than 10, maybe). Every new configuration of hardware or parameters will require a whole new 1Gb executable. It's not practical to consider a BOINC system where almost every WU will require 5 x 1Gb downloads (1 for each result) to complete.

However, some portions of the simulation could be downloaded effectively if they can be separated from the monolithic executable currently used. Therein lies the problem. How does one separate certain computationally intensive portions from others such that the data can be split and recombined properly?



Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 11540 · Report as offensive     Reply Quote
Profile Paul D. Buck

Send message
Joined: 2 Sep 04
Posts: 545
Credit: 148,912
RAC: 0
Message 11546 - Posted: 2 Dec 2005, 9:58:21 UTC

Well, CPDN shows that this is not completely impractical.

The Result Data Files are only 300M zipped, but still ...

And the run times are about 100 days ...

So, these are indications that this is not out of the realm of possibilities.

*I*, for one, am most interested in those projects where actual, usable, science is being done. Hence my lowering resources to SETI@Home and LHC@Home (which are speculative exploration and engineering respectively). So, I don't see the issue unless the runs HAVE to be done within short timeframes.

So, the question is, are the Geant4 models a one time item? Or, are many of them to be run. If there are, and time is not a factor, and there is not enough super-computer time to do them all ... well, it is still a good fit for BOINC.

"Typical" PCs now come with 200G disk drives. More and more people have DSL or Cable modem connections. Main stream PCs are now mostly well over 2 GHz (in P4 terms) in processing power.
ID: 11546 · Report as offensive     Reply Quote
Profile Yeti
Volunteer moderator
Avatar

Send message
Joined: 2 Sep 04
Posts: 406
Credit: 96,116,916
RAC: 14,701
Message 11574 - Posted: 5 Dec 2005, 23:03:53 UTC

1 GB download for each WU will not be a real problem for many users, but it will be a problem for the project-servers and the needed bandwith ...

Yes, CPDN is dealing with big WUs, but, they take 3 or more weeks to crunch; this saves bandwith again.



Supporting BOINC, a great concept !
ID: 11574 · Report as offensive     Reply Quote
Henry Nebrensky

Send message
Joined: 13 Jul 05
Posts: 102
Credit: 11,836,893
RAC: 24,225
Message 11625 - Posted: 8 Dec 2005, 17:27:27 UTC - in response to Message 11450.  
Last modified: 8 Dec 2005, 17:29:04 UTC

Well the software has been ported. The next step is taking one of the models probably the modelling of the Atlas detector and then decoupling the Geant 4 bit from the rest of the Atlas simulation...


Is this Atlas / BOINC work related to HEP@home, or has the latter died off?

Henry
ID: 11625 · Report as offensive     Reply Quote
Gaspode the UnDressed

Send message
Joined: 1 Sep 04
Posts: 506
Credit: 118,619
RAC: 0
Message 11626 - Posted: 8 Dec 2005, 19:53:33 UTC - in response to Message 11625.  

Well the software has been ported. The next step is taking one of the models probably the modelling of the Atlas detector and then decoupling the Geant 4 bit from the rest of the Atlas simulation...


Is this Atlas / BOINC work related to HEP@home, or has the latter died off?

Henry


Looking at the papers on the CHEP '04 site I'd guess that the whole HEP@Home project was an academic view of the requirments for processing data 'in situ', and wold require a modified BOINC system to work, rather than a new app for use in a BOINC environment. It wouldn't surprise me if you have to wait for CHEP '06 for the next update on HEP@Home. Have a nice time in Mumbai!



Gaspode the UnDressed
http://www.littlevale.co.uk
ID: 11626 · Report as offensive     Reply Quote

Message boards : Number crunching : What is happening with Geant4?


©2020 CERN