Meeting 01 31 2017 - ATLAS-Titan/misc GitHub Wiki

Meeting 01/31/2017

Present: Matteo, Sergey and Alessio

Agenda:

  1. Discuss Matteo's questions for Sergey.

Questions:

  1. Does each job of a PANDA Server (and therefore all the tasks of that specific job) belong to the same โ€˜userโ€™/VO? Does a PANDA Pilot pull jobs/tasks from a PANDA Server or does a PANDA server push jobs/tasks to a PANDA Pilot?

  2. Does a PANDA Pilot pull jobs or tasks from a PANDA Server? Are the jobs within a PANDA Pilot the same jobs (i.e., the same container of the same tasks) in the PANDA Server?

  3. Do we have 16 PANDA Pilots on a Titan DTN (is this the right acronym?) node? How many DTN nodes?

  4. Can the events packaged into a PBS job belong to different submissions, i.e., different jobs, i.e., different.

  5. Do we have the distribution of the time taken on titan to process 1 event of the Monte Carlo workflow?

Answers:

  1. Same user task->job on the server.

  2. Panda pilot is a edge service on titan, event-driven by backfill availability. When available, it DOES THE PULL from the panda server.

  3. Currently 20, thinking to go higher, on 4 DTN (Data Transfer Node), 5 pilots for DTN, DTN that enable interactive data transfer. New DTN farm is being rolled on. Our set up will change. More node, more pilots per node. This is bound by DTN performance, overloading DTN (currently absence of data on why these nodes are overladed. Network/memory should not be a problem, it is probably purely CPU because of tools used to move files, e.g., GridFTP.)

  4. They do belong to different tasks.

  5. To be added to Section VI about performance. See presentation: https://drive.google.com/drive/u/0/folders/0BxWZQExiTsWNRTdMQXN4a1g3bHc

TO DO: