sPHENIX tracking discussion

[Anthony Frawley <afrawley AT fsu.edu>]

Here are the minutes from today's tracking meeting. The agenda page is:

https://indico.bnl.gov/event/4359/

Cheers

Tony

------------------------

At the Feb 20 meeting we agreed that we should have a base class (inheriting from SubSysreco) for all reconstruction modules. The base class will contain all of the common parts (e.g. common node tree pointers) that do not need to be changed whenever we change the algorithm.

Today Haiwang showed some slides (see agenda page) on a prototype design of five base classes, and showed how existing and planned modules would relate to them.
He also showed which input and output pointers to node tree objects would be used in each base class, and showed a prototype container for association information.

Slide 3: Christof started a discussion about our need for much faster code. ALICE can track in less than a second, so it should be possible.
It is relevant here because it will certainly impose requirements on our storage objects.
ALICE breaks up the TPC into small segments and uses cellular automata to process them in parallel.
This will need appropriate packaging of data - need to go away from one big cluster map, for example.
Christof wants to try to get the ATLAS tracking code, which is becoming open source about now.
Sookhyun is interested in looking at the ATLAS threaded code. Marzia is working on ATLAS, will ask her if she can look at the code.

Slide 3 (common pointers): The base class supplies pointers to the node tree objects to the derived class
Sookhyun and Haiwang will talk more offline about details of these.

We are presently having a resource crunch due to data production for QM2018. But we have our own deadline - the CD1 review - and we had a discussion about how to deal with that short term.
We should focus on getting the memory down for now, since that enables use of the more numerous low memory Condor slots.
Truth matching takes a lot of time, and this is helped greatly by evaluating embedded tracks only.
Haiwang will do some memory profiling for TPC cells with central + 200 kHz events.
Darren's new storage scheme may help a lot - having one cluster object per detector element, not one cluster object per cluster will greatly reduce overhead.

Darren:
The pull request for TrkBase is still waiting on Chris for approval. He wanted some changes, did not say what they are.
Jin: Chris would like to change the coding style to better match the sPHENIX coding style. Minor changes. Chris is working on a coding style guide.
Darren said no progress on the association bits yet for digitized hits to cluster, it is on his TTD list.
Test beam last week. The association for the MVTX was done with the new TrkBase, worked very well.
Darren will ping Chris again. Merging TrkBase is not holding anyone up yet, still a lot of work before implementing. Darren is working on association modules while waiting on Chris.

For Darren, the next big thing is a geometry object for the MVTX.
The generic idea is to have a way to have parameter input from DB/file allowing easy configuration of the ladders in MVTX and INTT. Can be used during reco, to do alignment studies, etc.