Difference: ProductionHowto (14 vs. 15)

Revision 152013-03-18 - DanProtopopescu

Line: 1 to 1
 
META TOPICPARENT name="WebHome"

NA62 Monte Carlo Production Howto

Line: 20 to 20
 
Changed:
<
<
The aim is to maintain the production rate at its maximum (whatever that is, depending on the resources available) and for this the person on shift must submit new jobs when the number of waiting and running jobs is low. How is low defined ? We should have more than 200 jobs RUNNING at all times and about 50 SCHEDULED.
>
>
The aim is to maintain the production rate at its maximum (whatever that is, depending on the resources available) and for this the person on shift must submit new jobs when the number of waiting and running jobs is low. How is low defined ? We should have more than 200 jobs RUNNING at all times and about 50 SCHEDULED, but not more than 100 jobs in waiting states.
 ALERT! Please note that these numbers will change when new resources are added. Check this number at the beginning of each shift.

Job submissions in production mode are done via the Scripter interface, as explained below.

Line: 36 to 36
 This is an HTML form with many input filelds, most of them self-explanatory. The pre-filled values are inherited from the previous submission (which could have been a test job for example), so you must check that they fit the production round you are managing.
Deleted:
<
<
ALERT! There is now a simpler, production-only scripter interface. Use that for production! It has less settings than the one described below.
 Here is how the scripter interface looks like:
Changed:
<
<
ScripterUI.001.png
>
>
ScripterUI2.png
  Description of the form fields:
Changed:
<
<
  1. Description - this is a short description of the job. It must contain the production tag (get it from here if unsure), and it must contain the keyword "production". Should not contain quotes or any other non-text characters. Leave out the tag and replace production with test if you want to submit a few test jobs.
  2. Run interval - the start run is pre-filled with the next available run number (from the DB). Choose the upper limit such that you submit not more than 100 jobs at a time. Do not use job cloning for production at the moment.
  3. Number of events - this is the number of events per job (run). We aim to keep the job runtime below 20 hours, so for channel 1 an r188 that means 1500 events per job, for r193 and channel 1 the optimum would be 6000 events etc. For other channels it could be more or less than that. Check previous production jobs to find the optimal number of events. Leave the random seed as it is, because it will be set automatically for each run.
  4. MC software version - you must use the latest software version (check here if unsure), and make sure the wrapper script supports this version. Take a look at the scripts of previous jobs to make sure (click here, select the last production job, then click the corresponding green box link on the "Exe" column and check the line vers="v?"). There is a grid "version" for each installed software revision (e.g. v6/r188, v7/r193), see this wiki.
  5. Decay type - is the reaction channel to simulate. Choose from the drop-down menu and make sure it corresponds to the current production tag and description (check here if unsure). Leave default values for the remaining options, unless instructed otherwise.
>
>
  1. Choose description - this is now a drop down menu, containing items from the actual production schedule
  2. Run numbers - the start run is pre-filled with the next available run number (from the DB). Choose the upper limit such that you submit not more than 100 jobs at a time.
  3. Number of events - this is the number of events per job (run). We aim to keep the job runtime below 12 hours, so for channel 10 and v9/r261 that means 6000 events per job. For other channels you would have to calculate an optimum. How to do this: submit a 300-500 events job; when it finishes with output saved locally, go to the jobs table, click the "Submission Date/Time" cell to expand the row and get detailed info. Find in there the "events per second" figure. Calculate how many events (in multiples of 1000) we would be able to run and still keep the total runtime below 12 hours. Check previous production jobs as well.
  4. Leave the random seed as it is, because it will be set automatically for each run.
  5. MC software version - you must use the latest software version (check here if unsure). Take a look at the scripts of previous jobs to make (click sure. There is a grid "version" for each installed software revision (e.g. v6/r188, v7/r193, v9/r261), see this wiki.
  6. Radiative corrections, Disable detector(s), Disable Cherenkov - leave default value here ("off", "none,none"), unless instructed otherwise.
 
  1. Destination - tick here only the sites that have the chosen MC software version installed. Check this table to make sure. Check the jobs history to detect any problems at sites (e.g. jobs consistently finishing early, or going to status CLEARED without registering any output). ALERT! If jobs fail at a site, uncheck it here, notify the site admin and add a comment in the logbook!
Changed:
<
<
  1. Executable - automatically selected now - this is the name of the wrapper script that is executed on the worker node. It checks if the software is installed, runs the actual MC simulation, registers the output and triggers the FTS transfer(s). In single jobs mode, you can display commented lines from scripts in case you would like to check extra settings, comments etc. For production, leave this unchecked.
  2. User and password - for multiple job submissions, you need to tick the "Write scripts to disk" checkbox, and introduce your uid and password for this interface. ALERT! You must have registered and your credentials must have been validated for this to work. In single job mode, uid and password are not needed, since you will have to submit the (test) job with your credentials from your UI.
>
>
  1. In single jobs mode, you can display commented lines from scripts in case you would like to check extra settings, comments etc. For production, leave this unchecked.
  2. Write scripts to disk/User and password - for multiple job submissions, you need to tick the "Write scripts to disk" checkbox, and introduce your uid and password for this interface. ALERT! You must have registered and your credentials must have been validated for this to work. In single job mode, uid and password are not needed, since you will have to submit the (test) job with your credentials from your UI.
 
  1. Click Prepare, and you are taken to a new page. ALERT! If the page says "There are scheduled submissions in there. Please try again in 5 minutes" it means that you have (or someone else has) just scheduled another batch of jobs and you have to wait for these to be actually submitted, else the scripts may be overwritten - with unpredictable results.

Multiple Submissions

Line: 94 to 93
 
META FILEATTACHMENT attachment="ScripterUI2.002.png" attr="h" comment="Submission confirmation screen shot" date="1350047609" name="ScripterUI2.002.png" path="ScripterUI2.002.png" size="158051" stream="ScripterUI2.002.png" tmpFilename="/usr/tmp/CGItemp27615" user="DanProtopopescu" version="1"
META FILEATTACHMENT attachment="na62-button-apple-small.png" attr="h" comment="Webapp icon" date="1350122891" name="na62-button-apple-small.png" path="na62-button-apple-small.png" size="10818" stream="na62-button-apple-small.png" tmpFilename="/usr/tmp/CGItemp7460" user="DanProtopopescu" version="1"
META FILEATTACHMENT attachment="add-to-home-screen.png" attr="h" comment="Add to home screen" date="1350324838" name="add-to-home-screen.png" path="add-to-home-screen.png" size="25211" stream="add-to-home-screen.png" tmpFilename="/usr/tmp/CGItemp7754" user="DanProtopopescu" version="1"
Added:
>
>
META FILEATTACHMENT attachment="ScripterUI2.png" attr="" comment="Scripter UI" date="1363599975" name="ScripterUI2.png" path="ScripterUI2.png" size="86041" stream="ScripterUI2.png" tmpFilename="/usr/tmp/CGItemp8739" user="DanProtopopescu" version="1"
 
This site is powered by the TWiki collaboration platform Powered by PerlCopyright © 2008-2020 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback