OWG Meeting 25-Apr-2012

From GlueXWiki
Revision as of 14:58, 26 April 2012 by Wolin (Talk | contribs)

Jump to: navigation, search

Agenda

  • Announcements
    • Run-around at 3pm
  • Review of minutes from 11-Apr-2012 meeting
  • Mini-Hbook - Sergey B
  • Hall D Component Inventory Database
  • Counting house - Hovanes
  • TDR - Dave L, Elliott
  • 12GeV planning lines - David, Hovanes, Elliott
  • Elog and Accelerator division projects - Graham
  • DAQ group - Bryan, Dave A, Graham
    • CODA3, EVIO, ECS/CSS, raw event file, disentangler, readout list
  • EPICS - Hovanes, Vanik
    • transfer some projects to Accelerator Div.
  • Controls - Elliott
  • Electronics, trigger and DAQ status reports - Chris, Alex, Dave, Fernando, Ben, Ed, William, Bryan
  • Mantis Task Tracker
  • Midas Elog


  • Calibration/Alignment issues?
  • Computer system manager


Time/Location

1:30 PM Wed 25-Apr-2012 CC F326


Announcements

Next Meeting

1:30 PM Wed 9-May-2012 CC F326


Minutes

Present: Elliott W, Simon T, Bryan M, Sergey B, Hovanes E, Beni Z, Eugene C.


This was a short meeting due to the annual JLab Run-a-Round at 3pm. We broke at 2:30pm. Most of the time was taken up hearing from Sergey Boiarinov about the mini-Hbook system he implemented for the CLAS online. After that we had cursory reports about a few topics.


mini-Hbook

Sergey created this facility in 2004 to gather information about primary and secondary readout list execution times in the ROCS. Histograms are created and filled in the ROCS, then transmitted to display programs via the event stream (ET system) every 100k events (every few secs or so). The display programs use config files to tell them which histograms to extract from the data stream, how to display them, whether and how to sum them, etc. The origianal version worked well and he expanded its use somewhat over the years.

Only a few Hbook-like API calls are implemented, just the ones needed to book and fill 1-D and 2-D histograms. The display programs accept the mini-Hbook histograms and convert them to real Root histograms, then use the full power of Root to display and analyze them. In this way the memory/processing footprint in the ROC is minimized, with no dependence on large libraries, version incompatibilities, etc. He strongly recommends against using Root itself in the front-ends.

Currently the histograms are serialized to BOS format prior to injection into the CLAS data stream. For CLAS12 they will be converted to EVIO format. Once serialized the histograms can be transmitted in a number of ways, including via the DAQ cMsg system. Note that an ongoing debate concerns whether to transfer meta-data, such as histograms, in-band via the ET system or out-of-band via another system. Probably it's must a matter of taste. Sergey prefers in-band but recognizes that other mechanisms should work just as well.

Serialization is optimized to send binned data or sparse data, whichever is most efficient. Many large histograms have sparse data so this is an important optimization. Underflows, overflows and some other statistical data are accumulated.

The CLAS12 offline may use this facility in their CLARA environment. Here a large amount of data gets transferred over the network during event processing, and minimizing the memory footprint of histogram data may be critical.

Sergey's library is currently only available in C, but it is possible to wrap the C library via SWIG and thus create API's in Python, Java, etc.

Sergey and Elliott will place the mini-Hbook library in the 12GeV repository so Hall D can test it.

Retrieved from "https://halldweb1.jlab.org/wiki/index.php?title=OWG_Meeting_25-Apr-2012&oldid=36545"