MUIOO/LVL2: Run-4

Goals

To have a module run fast in LVL-2 (Run-4), which should work close to/as well as/ the offline MUID road finder. Instead of doing the fitting in one orientation at a time, try to do it for both and beat down the combinatorics up-front. The algorithm is basically a copy of the MUTR/MUTOO stub finder.

Status and to-do list

Number of roads reconstructed, and reconstruction times for different filetypes and algorithms have been studied. So far, only simulation files have been looked at. Obviously want to test next with real data PRDFs from Run-2 and Run-3. Also, for the simulations: should try to quantify accuracies of the various algorithms, and efficiencies and ghosting. Perhaps best done with signals embedded in data(?).

Algorithms

The algorithms used in muioo land, are:

mMuiRoadFinder1 - which should be roughly equivalent to the standard mui first pass road finder. I'll refer to this with the abbreviation Orig.
mMuiFastRoadFinder - which should mimic the performance of the Local-Level-1. I'll refer to this with the abbreviation LL1.
mMuiFindRoad - this is the latest addition; algorithm a la stub finder. I'll refer to this with the abbreviation New. It is not quite run as the Orig version; e.g. the seeding order is basically front-to-back or back-to-front (reverse mode). No grouping/golden selection is done; if two roads share too many hits, the road with smallest number of hits is removed.

Filetypes

The types that have been studied so far, are:

Simulations:
Single muons, from Chun, with high momenta.
J/Psi - a signal sample, from Sean. Note that not all are reconstructable.
d-Au - Hijing min bias.
Au-Au - Hijing min bias.

Nroad Comparisons

A re-arranged summary is given here.

Filetype Single J/Psi d-Au Au-Au
Orig 1.43 +- 1.49 2.17 +- 1.27 0.01 +- 0.10 0.49 +- 1.88
LL1 1.00 +- 0.00 3.32 +- 1.01 0.00 +- 0.00 0.99 +- 3.82
New 1.23 +- 0.91 1.94 +- 0.85 0.02 +- 0.14 0.04 +- 0.24

Time Comparisons

In abscence of a good comparative timer (yes, I'll try to use the mutoo timing stuff instead), I just looked at the ratio of time spent in the road-finder module compared with the cluster-finder module, as reported by jprof.
(The road-finder module changes between the algorithms, the cluster-finder module does not, so we can use the cluster-finder for normalization..)

Filetype Single J/Psi d-Au Au-Au
Orig 56 / 24 = 2.33 114 / 47 = 2.43 14 / 13 = 1.08 31 / 21 = 1.48
LL1 11 / 20 = 0.55 6 / 47 = 0.13 5 / 10 = 0.50 1 / 23 = 0.04
New 57 / 23 = 2.48 156 / 42 = 3.71 9 / 16 = 0.56 27 / 26 = 1.04

Location of input files; code and macros

SlowSim DSTs:
/phenix/data22/silvermy/sandbox/response_r

DSTs (& log files):
/phenix/data22/silvermy/sandbox/mutoo/analysis/sim_to_dst

Ntuples (& log files):
/phenix/data22/silvermy/sandbox/mutoo/analysis/dst_to_ntuple



silvermy@lanl.gov 2003

Last modified: Sunday, 31-Aug-2003 18:35:30 EDT