the data from the front-end cards and transmit the sub event to readout buffer via optical S-Link at a speed of 160 MByte/s. The data coming from each link are stored in 512 MByte spill buffer cards in the central Read-Out Buffers (ROBs).
The data are distributed from the ROBs to twelveevent−builders work-ing in parallel, where they are combined to global events. In 2004 the total data transmitted during the spill to the readout buffers correspond to 230 MByte/s. Fi-nally these data are recorded on tape remotely at the CERN central data recording facility located in the CERN computer center.
3.9 Data-analysis at COMPASS
A huge amount of data (nearly about 350 TB per year) is collected by the exper-iment. This requires the availability of sufficient computing power to reconstruct the events at a rate comparable to the data acquisition rate. The demanded CPU power is provided by 200 Linux Dual-CPU PCs at CERN. First step in handling of the data is called theProduction process, where the raw data are used to reproduce particles and their tracks. After this process is completed, the data is ready to use for different physics analyses.
Raw data from COMPASS experiment are stored in the Central Data Recording (CDR) system at the end of every run and transferred to the COMPASS Comput-ing Farm (CCF) where the data are stored for a certain period of time before beComput-ing written to tape. Data access from tape is established through the CASTOR9 sys-tem where the user is provided with a directory structure and commands for writing, reading, opening and closing of the files. When a file is being requested by an user, CASTOR downloads a copy from the tape onto a local disc for access.
3.9.1 Event reconstruction
The Event reconstruction is performed by a fully object oriented program called CORAL10. This is designed with a modular architecture with collection of class li-braries written in C++ programming language. A schematic representation of the reconstruction program is shown in Fig. 3.14.
The input given to the reconstruction procedure is either the raw data collected by the experiment or the output from Monte Carlo simulation software.
Only the reconstruction of real data is explained here. Before the raw information from the detectors is sent to the track finding algorithm, two initial processing phases
9CERNAdvancedSTORage
10COMPASS Reconstruction andAnaLysis
3.9. DATA-ANALYSIS AT COMPASS 47
Figure 3.14: Schematic representation of the COMPASS reconstruction software.
are required. Decoding :In this first phase, the information from the fired detector is extracted from the raw data. Clustering : In the second phase, the information from the detector channels that have fired by the same particle are grouped together.
Finally the different steps of the data reconstruction are performed, like track and vertex reconstruction and also the hadron identification by RICH reconstruction which are explained in the following part.
Track reconstruction: The spectrometer is divided into three regions by the two spectrometer magnets SM1 and SM2 respectively. This is done by the track reconstruction algorithm named TRAFFIC/TRAFFDIC. The three phases of reconstruction corresponds to: a) pattern recognition, where, the track segments are identified in the various zones of the spectrometer then, b) the track segments from several distant zones are connected to build full tracks (Bridging) and finally c) the best estimators for the parameters of the reconstructed tracks are computed (Fitting).
3.9. DATA-ANALYSIS AT COMPASS 48
Vertex reconstruction: Various tracks that have a common point sug-gest the existence of an interaction point or a vertex. Here the aim is to get the best estimator of the three coordinates of the vertex position from each track that is assumed to originate from there. All tracks surviving the initial phase are used to estimate the parameters of the vertex and the relative χ2 fit is performed, which acts as a measure of probability of a particular vertex. The approximation of the primary vertex is possible by computing the average Point Of Closest Approach (POCA) between one beam track and all possible outgoing tracks.
Particle identification: The particle identification is performed inside CORAL by a package named RICHONE. The raw information from the RICH de-tector provides the coordinates of the photon dede-tector pads with a signal above threshold. This information combined with the particle trajectory from track re-construction is used to calculate the probability that the particle is of a particular type.
After the reconstruction phase, the information such as track parameters, vertices, calorimeter clusters, PID probabilities, detector hit patterns, etc are stored into output ROOT trees called as mini Data Summary Tapes (mDST). The data reduction factor between the input raw data and the output mDST is about 100.
3.9.2 PHAST
PHAST11 [75] is an internally developed program. It is the main tool for physics data analysis at COMPASS. This program reads in all objects from the mDSTs. It provides dedicated classes and functions based on the CERN ROOT packages [76].
Using these common routines, various physics analysis can be performed. The data analysis explained in Chapter 4 was performed with PHAST.
11PHysics AnalysisSoftware andTools
Chapter 4
Extraction of transverse spin asymmetries at COMPASS
“What we observe is not nature itself, but nature exposed to our method of questioning.”
Werner Heisenberg