Message boards : Cafe LHC : Status Dec 2005.....
Message board moderation

To post messages, you must log in.

Profile The Gas Giant

Send message
Joined: 2 Sep 04
Posts: 309
Credit: 715,258
RAC: 0
Message 21477 - Posted: 16 Aug 2009, 11:10:03 UTC

In cleaning up an old hard disk I came across the following article regarding the status BOINC at LHC in Dec 2005...yes 2005!

Status of the IT BOINC Project
Ignacio Reguero 20 December 2005

Here follows a report on the current status of BOINC project. In summary we could say that we are combining a successful LHC@HOME Sixtrack production campaign with work to rationalize the management of BOINC as a service while we explore other possible BOINC applications of interest to the LHC and HEP community.

Over 73 million BOINC credits have been produced with over 21 thousand persons subscribed to donate CPU. This means around 250 KSfp2K*year delivered to AB for the Beam to Beam tracking studies.
As reported by Werner Herr, these results have been presented by Dobrin Kaltchev in an AB internal meeting last Friday. The presentation was well received and it was decided to continue with another massive study after Christmas, using a slightly different set of parameters. The foreseen study will need about the same computing resources as the one just completed, maybe more, probably not less. The results of the previous study will be presented at the forthcoming Particle Accelerator Conference 2006 in Edinburgh.
Our only worry here is that if AB does not present a steady flow of load to LHC@HOME the users that donate their CPU time may be disappointed and look for other projects. We are explaining this danger to AB, we are also considering filling the gaps with alternative applications.

Server Management
Philippe Defert has produced RPMs to allow the Quattorization of BOINC Servers. This is a necessary step to rationalize the management of our BOINC servers.We plan to use it for the internal deployment.
The work to enable the automatic generation of these RPMs is being fed back to the BOINC project.

This is the recycling of 100 obsolete farm PCs as BOINC nodes. We produced an RPM that was integrated in a Quattor profile to automate the conversion of the nodes. This has been a great success. These nodes quickly became a top LHC@HOME contributor. They also give as a lot of flexibility to pursue metascheduling studies and to pursue other BOINC applications without PR constraints.

Internal Deployment on Windows Desktops
Although the proposal to deploy “Screen Saver” software to profit from unused CPU cycles of CERN Windows desktops was approved by the Executive Board in the summer it did not take place due to technical problems.
We were proposing to deploy the CPSS CERN homegrown product for this purpose rather than BOINC but as discussed in the department program of work, it is now difficult to defend using both CPSS and BOINC for the same purpose and there are concerns about the cost of maintaining CPSS in the longer term. So after discussion with Eric and Andreas we proposed to verify that an internal BOINC server can be made functionally equivalent to CPSS for usage on CERN Internal Windows desktops. In order to do so Christian has configured the BOINC server that we made available to Eric. If no major problem is found we propose to move forward with deployment beginning of 2006. We propose Philippe Defert to take responsibility for this deployment working together with the Windows team.

Project accepted by the DG provided that we do not spend CERN funds. The goal is to run Malaria epidemiologic simulation.
The project is running on recuperated obsolete disk servers. Currently have two well tested servers and two more being tested.
Alpha test successfully done with 26 experienced users. No interruption for three weeks. Automated job generation used. Have to test scalability on the database. This is interesting for other BOINC projects.
Beta test will be done when Jacques is back later with a few thousand users.
Web pages and definitive name are currently being set up.
Geneva International Academic Network and University of Geneva are supporting the project. Two African students came to CERN during the summer funded by Geneva International Academic Network. They did quite a good job.
Once development is finished the server is to be transferred to the University of Geneva. The University of Dakar has also shown interest to host a server.
The project has been a great PR success. It has been presented at the WSIS in Tunis, where Ben was also interviewed by BBC World Service radio for a broadcast made on November 21st and now available as a podcast.
The project was presented, together with IBM\\\'s World Community Grid, on 8th December in the First Tuesday public event at CERN.

Kickstart workshop held in CIEMAT in Madrid 14-17 November as CIEMAT is coordinating all technology projects for Extremadura.
BOINC servers and clients have been installed and configured in their Linux version (Debian).
The seminar material has been fed back to the BOINC project (and posted in the BOINC project site).
Potentially 60K PCs owned by the Extremadura administration are available. We are now expecting pilot with a considerable number of PCs beginning of 2006.
We are also expecting one FTE split in two 6 months visitors at CERN sometime on 2006.
A Computing Seminar was held on the 8th December with the material produced for the workshop.

Event Generators
As we got several indications that this type of software could be well adapted to BOINC, Juan and Ignacio are working on a demonstration of the Pythia software under BOINC.
We contacted the GENSER project leader (Paolo Bartalini) who said that he would keep in mind our proposal.
We are planning to contact directly the people that are to run a \\\"event production facility\\\" for the GENSER LCG project to check whether the would be interested in BOINC.

To calculate Feynman diagrams to obtain matrix parameters to improve accuracy of event Generators.
Proposing collaboration with ACPP with Japanese and Russian groups
To use Compep, GRACE and Vegas(Multidimensional integration) software.
Proposing to split the phase space and distribute it among clients in iterative processes. This poses specific requirements to BOINC
Denis proposed to do a Monarch simulation model. This is important for traffic and capacity planning
We are now waiting for the formal project proposal to be finished.
Eric is interested to help with the numerical aspects.
John Ellis has shown support for the project.
We need to show other support the project has in the HEP (and LHC) community to our management.

In contact with the GEANT4 team, Juan Antonio and Ignacio demonstrated GEANT4 release test software from them. The software was ported to BOINC and to Windows without any major problem. This work has been documented in Twiki and presented it to GEANT4 Workshop.
We are now looking for other (more realistic) use cases. We have been proposed to port the ATLFAST model of the ATLAS Fast Simulation. We are looking for how to disentangle the software from the rest of the ATLAS software before trying a port.
We would also welcome other suggestions for models.

Rob Veenhof is the author of the Garfield software package that is included in the LCG project. He has proposed us to port his software to BOINC environment to solve the CPU requirements of the model in use for the ATLAS detector so ATLAS would be very interested in using BOINC for this case as quickly as possible.
The specific program to run (Magboltz) is a relatively small and self contained piece of FORTRAN. We plan to demonstrate a BOINC port in January and decide for production after we are sure there is no obstacle.
Here follow some technical details from Rob

• This software computes transport parameters in gas mixtures.
• One has to run Magboltz for each (E,B,angle) triplet that may be relevant.Each point, at good precision, takes between 30 mins and 5 hours (order of magnitude). One would typically want to have 50 points in E, 20 points in B and 20 points in angle - but more wouldn\\\'t hurt. This needs to be done for several mixtures: for a start, one could think of 20 Ar/CO2 mixtures, 20 Ar/CH4 mixtures and some Ne/CO2 as well. Lots of other things could be envisaged at a later stage.
• For each gas, this comes to 2-40,000 hours of CPU time and we can easily find some 20 gas mixtures. So, filling 0.5 to 1 Mh of CPU hours isn\\\'t hard. Beyond that, one has to think.
• The Garfield interface ensures that each point can be run individually, or grouped.
• It wouldn\\\'t be hard for me to create an executable that is self-contained in the sense that it needs only an (excellent) random number generator, but no other external routines and it wouldn\\\'t generate graphics either. For monitoring, it would be easier to have a somewhat larger thing with graphics.

(S@H1 8888)
ID: 21477 · Report as offensive     Reply Quote

Message boards : Cafe LHC : Status Dec 2005.....

©2024 CERN