GlueX Data Challenge Meeting, November 19, 2012
GlueX Data Challenge Meeting
Monday, November 19, 2012
1:30 pm, EST
JLab: CEBAF Center, F326/327
- Minutes from last time
- Conditions for DC-1
- run numbers
- random number seeds
- file naming convention
- events per file
- CKOV in control.in
- JLab Farm Mini-Challenge Status
- 1000×100-k-event jobs over weekend
- combined simulation and reconstruction
- single-threaded, 10 hours per job, 224 jobs at a time
- Grid status
- CMU status
- start date
- New action items
To connect from the outside:
- Call ESNET Number 8542553 (this is the preferred connection method).
- Phone: (should not be needed)
- +1-866-740-1260 : US and Canada
- +1-303-248-0285 : International
- then use participant code: 3421244# (the # is needed when using the phone)
- or www.readytalk.com
- then type access code 3421244 into "join a meeting" (you need java plugin)
- CMU: Paul Mattione, Curtis Meyer
- JLab: Mark Ito (chair), David Lawrence, Dmitry Romanov, Simon Taylor
- UConn: Richard Jones
Review of Minutes from the Last Meeting
Data Challenge Configuration
Reviewed the conditions page and discussed some details of the challenge:
- run number all files will be a part of run 9000
- file naming convention Mark will propose one
- random number seed We will make sure that the seeds used for bggen, hdgeant, and mcsmear are all recorded in the output file. David will have to modify his previous implementation of this feature to accommodate all of these seeds. Richard propsed (some time ago) a web sevice that would distribute and record seeds. Although an attractive idea, we probably not get to that for this challenge. Mark will propose the seed generation scheme.
- number of events per file Mark will make a proposal.
- Cerenkov processes We will not turn on Cerenkov radiation. There are no detector element for which this physics process makes a difference in the artificial world of HDGeant.
Mark ran a mini-data challenge over the weekend of 100 million events (1000 jobs, 100 k events each) using the configuration on the web page. Simulation and reconstruction were done in the same job and only the REST format data were kept. 224 jobs ran in parallel and each one took about 10 hours of CPU time and used about 1 GB of memory. More simultaneous jobs should be possible by asking for less than the 4 GB he requested per job. The jobs ran single threaded.
Paul has successfully built the version of sim-recon called out on the conditions page. He will take a look at running the binaries.
After quick look at the conditions page, richard thinks that it conatins all of the information he needes to configure gridmake to build the binaries on the remote grid nodes.
We agreed to try and start production by next Wednesday, November 28. We can use the Offline software Meeting that day for any last minute adjustments.