CWG13 meeting

Friday 11 Apr 2014, 11:00 → 12:35 Europe/Zurich

301/R-007 (CERN)

CWG13 regular bi-weekly meeting.

CWG13 Meeting, 11/04/2014 11:00 301-R-007
 
Present: Andrei, Barth, Charis, Peter, Predrag, Raffaele, Ruben, Sylvain, Vasco
Remote: Ken, Matthias, Mohammad, people from GSI
 
Data model (Andrei) + discussion during the presentation:

• There is no implementation of the model for the moment;
• The shared memory representation restricts the data model to C structures, eventually inside a C++ class;
• The data are both raw and results of processing, very similar to what we have now in HLT;
• It is not granted that we will have event (interaction) building in O2, we may need offline processing to attach the tracks and clusters to primary vertex and to define the interaction. Only after this step is completed we can do analysis;
• The compressed time frame will not be modified, the new information will be added;
• The multiple data header adds description about the new types of information and permits fast navigation;
• The EPN waits for HBE summary. Since the reconstruction on each EPN is much slower, the latency is not important;
• The permanent extensions contain i.e. check sums;
• Why do we send time frames before they are complete?
  • This correspond to different steps of processing;
  • We may need intermediate checks inside the time frame.
• The calibration is associated with the time frame, but the algorithm may take longer, so we probably need more than one data stream from FLP to EPN;
  • Publish the calibration information in a separate stream, correlated/linked, but not integrated within the time frame;
• EPNs will need the full history for the space charge corrections;
• A possible data structure of the CWG13 repository:
  • Base: contains i.e. the descriptions of the data structures, base classes, etc.;
  • Common: common algorithms;
  • FLP specific code;
  • EPN specific code.

 
Action list:

• Common meeting with CWG3 on the calibration data flow (16/04/2014);
• Start the development of the data model headers.

11:00 → 11:45 Data model

Speaker: Andrei Gheata (CERN)

Slides

• 20140411_cwg13.pdf
• Data_model_CWG13_kickoff.pdf
• Data_model_CWG13_kickoff.pptx

11:45 → 12:30 Ideas and discussion about the online part of the prototype

Slides CWG13_2014_04_11.pdf CWG13_2014_04_11.pptx

12:30 → 12:35 CWG3+CWG13 meeting, 16/04

Slides https://indico.cern.ch/event/314038/

Common CWG3 & CWG13 Meeting, 16/04/2014 14:00 60-2-023
 
Present (CWG13): Andrei, Charis, Mohammad, Peter, Raffaele, Ruben, Sylvain, Vasco
Remote (CWG13): Matthias, Ivana, Jens
 
Ideas on conditions DB and data flow (Raffaele) + discussion during the presentation:

• The current calibration produces about 10⁵ files/year, which represents 40 GB/year.
• One third of the files needed for the sim/rec of a given run are automatically updated (i.e. per-run validity, all other objects have long run-range validity).
• Relevant information specific to calibration types is: the validity period (frequence of updates), the statistics needed (how much time to achieve the calibration), and the size of the object (less important).
• Calibrations needed at run-time (for TPC): drift velocity, effective gain, space charge.
• At every instance the "average" space charge distortions correction map is rescaled by the integrated charge of previous 160ms, either from the multiplicity from the FIT or from TPC currents. FLPs associated with TPC can integrate this information and propagate it to the EPNs.
• We can associate the FLP calibration to the raw data since the FLPs can build the sliding window for the calibrations which need it.
• The upper estimate of calibration data rate: 5MB/5ms = 1GB/s. It corresponds to 5% increase with respect to the raw data stream (20GB/s from TPC). Some additional improvements would permit to achieve additional reduction by order of magnitude.
• Do the FLPs read calibration data? No, FLPs only add calibration data to the raw stream.
• DQM is consumer of calibration data.
• For each EPN we need a local copy of the ``static'' calibration information (at initialization, e.g. start of the fill, plus at regular intervals); this (periodic re-)initialization has to be coherent.
• Merging of calibration data among EPNs (similar to event building) requires a calibration dedicated EPN; it might also be needed to reduce the frequency of read/write OCDB accesses.
• The calorimeters are specific case, we have to take into account the 5 calibration steps that are present now.
• The DCS information – pressure, temperature, etc. is needed. One possibility to have it is to provide a DDL link from DCS and record this information in the raw stream.

 
Action list:

• Provide detailed block diagram for the calibration schema (CWG6);
• Provide data flow diagram for the calibration information (CWG6);
• Simulate the data flow that corresponds to calibration (CWG3).