Grid-version+testing+of+Cosmic+Ray+e-Lab

We have been asked by the I2U2 development community to test--promptly--a new version of the Cosmic Ray e-Lab which is capable of running in multiple environments: locally, on a "server farm" of dedicated computers, and on the grid (which right now is just a single additional machine, but will eventually be a world-wide network of computational resources. The purpose of the testing is both to see how the new code performs under pressure from multiple users running data-intensive jobs, and also to compare its various new modes of execution.

For this study, we will analyze four large data sets (each consisting of one month of data for each of the months of Sept thru Dec of 2005) that were recorded at Saint Joseph's High School. (Search for Detector ID 138, and look under Saint Joseph's High School to find the months in question. Choose "all" to select the entire month's data.) Each of these four data sets will be analyzed using five different analysis configurations: 1) VDS; 2) Swift Local; 3) Swift on the I2U2 Cluster; 4) Swift on the grid; 5) Swift in automatic mode (assigning the job to one of the first three Swift modes by some criteria embedded in the Swift code.) That leaves a total of 20 large runs: five for each of four data sets. We want to launch all 20 of these tests within roughly the same (15 minute) period of time. It will take a long time to run these tests, so you will have to submit the jobs on one day, and check back for results the next.

To partcipate, please do the following:

1. Sign up (using your initials) on one of the spots on the table below.

(To see a graphical overview of the way these analyses are performed, as well as a brief description of what each is for, see [|here].)
 * Execution Mode || September 05 || October 05 || November 05 || December 05 ||
 * VDS || DJ || EL || JV || DH ||
 * Swift Local || PF || CC || AP || KH ||
 * Swift I2U2 cluster || CR || SG || CM || CN ||
 * Swift Grid || MM || KS ||  ||   ||
 * Swift Automatic ||  ||   || DO,JK || DO,JK ||

2. Log in to [|this site] using the SJHSRC username and the PW I will give you in class. Navigate to the Data tab, select the flux study, locate detector 138 data for SJHS, run a flux study for your dataset in the execution mode you've chosen. (To select your mode of execution, use the "execution controls" under the "plot controls" on the staging page, the one that pops up immediately after you've chosen your data and hit the "run flux study" button.)

3. Queue the study, view the "analysis list" at the prompt, and record the time your study began. Wait until the study runs (perhaps overnight), and then record the time the study has stopped from that same list. (If you have to wait overnight, this list may be reset, and we'll have to get the data from the developers. Enter the total elapsed time (in minutes, rounding to .25 min = 15 sec) next to your initials in the table above.

4. Finally, in the discussion tab for this page, give a brief description of the execution mode you chose (using the graphical and tabular overview linked in #1, above), and for any other single mode of execution for your same data file. Which mode of execution took longer? Is that what you would have expected from the descriptions of the modes of execution you read?