70 likes | 163 Views
Reconstruction in JIM: status and plan. Gabriele Garzoglio Sep , 13 2004. The Goal. Submit a reconstruction job to the SAM-Grid infrastructure from any JIM client installation Execute the reconstruction job to any SAM-Grid execution site
E N D
Reconstruction in JIM: status and plan Gabriele Garzoglio Sep , 13 2004
The Goal • Submit a reconstruction job to the SAM-Grid infrastructure from any JIM client installation • Execute the reconstruction job to any SAM-Grid execution site • No d0 software release is assumed to be pre-installed at the site • Refine monitoring using the JIM information management infrastructure Gabriele Garzoglio
Status • Working on the infrastructure at the execution site • Iain has provided a tar bal of p14.06.02 that can be used together with mc_runjob to do reconstruction on a “bare” node • The tar ball has been tested interactively on a single local file and needs a few additions (mainly the RCP database). Gabriele Garzoglio
Short term plan • Store the tar ball (+ RCP db) into SAM • Write a jim-job-manager plug in, in order to submit the reco job to a batch system • Submission tests can be done at samgfarm • Eventually the submission will be triggered via the grid • A wrapper around the job will prepare the execution environment: • It will ask SAM for the tar ball, mc_runjob and an input file • It will generate the runjob macro Gabriele Garzoglio
Longer term plan • Enhance the jim-job-manager plug in to submit parallel batch jobs • Have mc_runjob talk to sam to get the input files (as opposed to doing it in the job preparation wrapper) and store the output • Write a client plug in to pass the user information to the jim-job-manager • Start testing submission over the grid Gabriele Garzoglio
Open questions 1 • What information does the client need to pass to the job? • input sam dataset • d0reco version • binary sam dataset • ??? • Is this information going to be encapsulated into a request system? Gabriele Garzoglio
Open questions 2 • What is the model of grid job submission? • Deal with the reco of 100 or 1,000 or 10,000 files as a single grid job? • What policy of local job submission should we enforce? • Submit as many local jobs as files in the dataset, processing only a file per job • Submit a fixed number of jobs that will process the whole dataset • ??? Gabriele Garzoglio