Agenda

Review minutes from May 20, 2009 Software meeting
Announcements
1. Special system user account at JLab: gluex
2. GPU discussion next time
New magnetic field maps: David
Bug fixes from Richard
Kalman Filter update (Simon)
Complexity growth in hdgeant: Richard
Work disk: Mark
Software tag policy
Action item review

Minutes

UConn: Richard Jones
JLab: Eugene Chudakov, Mark Ito (chair), David Lawrence, Sascha Somov, Simon Taylor, Beni Zihlmann

Review of Last Meeting's Minutes

We spent some time discussing Matt's suggestion (from two meetings ago, actually) of having all random-number related processes, such as smearing, occur in a stage before reconstruction and having the results of that stage written to disk. David pointed out that there is already a program, mcsmear that takes the hdgeant output and produces smeared hits in HDDM format, suitable for the reconstruction program. Right now only tracking is smeared but it would be straightforward to do smearing of the calorimeters as well.

Richard asked if the smearing used the truth information. David told us that that was done earlier but now the smearing is only applied to hits (i. e., to detector response information).

Announcements

Special system user account at JLab: gluex

Mark reminded us of the existence of this account. David reminded us that there are several cron jobs currently running under his personal account that would should be transferred to the gluex account.

GPU discussion next time

The often-postponed discussion of using GPU's for data analysis is tentatively planned for the next Offline Meeting. Sandy Philpott and Jie Chen of the JLab IT Division, Xipeng Shen of William and Mary, and Richard plan to participate.

New Magnetic Field Maps

David reviewed the new maps that we received recently from Floyd Martin. He has written a HOWTO on the procedure to convert Floyd's Excel files to our field map format. There he also describes how to produce a ROOT file of the map (Elton has been using these). To use them, the correct command-line argument needs to be supplied to the reconstruction program to shift the field by the appropriate amount in z. David will provide a table for now with the correct values, but in the future this should not be necessary when all useful maps have the same coordinate system.

Richard pointed out that the ability to apply a z-offset is provided in HDDS. This feature is not being exploited in the current code.

Bug Fixes from Richard

Richard reviewed his recent email.

missing CDC and FDC hits

This appears to be an old, well-known feature of GEANT3 when the step size is made small to account for thin layers of material. Due to an internal optimization strategy the answer to the question "have I crossed the boundary" can be answered in two different ways, both are used in some instances and if the answers conflict, swimming is halted(?). Richard put in a fix to catch these cases and allow swimming to continue. The change was made in gtnext(?). Richard encouraged us to test the code and confirm that the problem has been solved.

2 GB output file limit in HDGeant

This was fixed by an adjustment in the use of the XDR API. XDR is used to write machine-independent (endian-independent) binary data. In addition we need to compile our code with the gcc switch -D_FILE_OFFSET_BITS=64 thrown whenever we compile. David agreed to implement this change.

Kalman Filter Update

Simon showed further progress on his fitter. The two most recent changes are to do a wire-based fit before using drift information and to change the cut-off angle where he changes from a forward parametrization to a central parametrization. He showed big improvements in the formerly problematic areas around 20 to 40 degrees in polar angle. Resolution is generally slightly better than for the least-squares fit and in a few regions a lot better. There is still lots to do (in particular looking at the fitting efficiency) but the improvements were marked.

Complexity Growth in HDGeant

Richard noted that there have been several changes in HDGeant that have added dependence on other large packages (such as ROOT) and on our reconstruction libraries (such as tracking). He thought that if this could be avoided, it would make maintenance of HDGeant simpler (the "fewer moving parts" argument). Discussion focused on the the distance-of-closest approach algorithm in the tracking package. David introduced the algorithm used in tracking into HDGeant it to get consistency between Monte Carlo and reconstruction. He was not getting agreement with the straight-line-while-in-a-cell approximation formerly used in HDGeant.

We decided that David would back out the change in question and when he was done, Richard would take track curvature information into account when creating hits.

Deferred Agenda Items

Work disk information
Software tag policy

Action Items

We agreed that we should revive our regular review of old items and listing of new ones. Here are the items generated at this meeting:

Introduce calorimeter smearing into mcsmear.
- Tell Matt about our discussion -> Mark
- Consult with Matt on how this can be done -> David
Add -D_FILE_OFFSET_BITS=64 switch to gcc in makefiles -> David
Back out tracking doca algorithm from HDGeant -> David
Include "arc correction" in HDGeant straw hit generation -> Richard
Bring back the "Action Items" agenda item in future agendas -> Mark

Recorded by Marki 09:35, 19 June 2009 (EDT)

June 17, 2009 Software

Contents

Location

Telephone

Video Conferencing