DST Production and Skimming

The events accepted by the Level 4 software filter are reconstructed and the information is stored as DST. In this stage, raw data, the contents of which are direct logs from data acquisition devices, are converted into physics objects of 4-vectors of and . The most downstream of the reconstruction flow is the event classification and skimming. Here, all events are examined in such a way that certain selection criteria are applied to select events of our interest from a large number of background events. Once events are satisfied with our cuts, they are recorded into not only DST but also specific data files on the disk as skimmed data. In this way, physics events such as hadronic events and Bhabha candidates are stored. Based upon those data, detector calibrations are carried out in detail and the offline luminosity is computed. After all information is reconstructed, monitoring modules are called to keep our data quality reasonably high. Basic observables like the number of charged tracks reconstructed are booked and those histograms are checked by experts. A typical processing speed of DST production, including skimming part, is around 40 Hz, depending on beam conditions. In our computing machines, we can process about 80 $pb^{-1}$ of data a day when three CPU servers out of seven are allocated. Fig. shows a history of our event processing rate together with the data acquisition rate in the recent runs. Our offline processing has been well tracked to the online trigger rate with normally one day delay. The data size of DST is about 100 KB and 60 KB for hadronic and Bhabha events, respectively. The calibration constants, which are used to correct detector responses, are kept in (www.postgresql) database in our system. During the period of an experiment, detector raw responses can be shifted due to changes in environmental conditions. The energy loss measurements by CDC, for example, suffer from atmospheric pressure changes. After checking skimmed data created by DST production job, the detector constants are improved on the weekly basis to correct for this sort of effects. By getting updated constants, the re-processing has been made for hadronic events.

Figure: History of the event processing rate together with the data acquisition rate.

In physics analyses, one does not need complete information available in DST. Instead, easy access to experimental data is highly demanded because we must play with a large amount of beam data at the B-factory. For this purpose, minimal sets of DST (``mini-DST''), which is compact but sufficient to study physics channels, are created for all runs. The hadronic event size at this level is about 40 KB.