The amount of CPU resources needed for the event processing in the hadron program of COMPASS is estimated on the basis of the current experience with WA89 data. Table 3.19a shows the total number of events recorded and the processing time for one event.
Table 3.19: Comparison of CPU times and processing speeds.
The WA89 data was stored on 2000 Exabyte 8200 tapes and is processed on 36 DECStations 5260 and a 6 CPU AlphaServer 8200.
After the online filter stage COMPASS will write about 30 times more events per year than WA89. Although the event complexity in COMPASS will be higher than in WA89 we expect a gain in reconstruction speed of 1.5 due to much faster reconstruction algorithms available. This gain together with a projected increase in CPU performance of 2.5 ( e.g. Digitals Alpha ev7 announced for 1999) leads to the corresponding numbers for COMPASS listed in Table 3.19b
The total processing time of s for one year of data taking will lead to a total processing time of about 100 days on a farm equivalent to 40 fast processors. Multi-processor arrays with low cost CPUs will also be considered.
In the COMPASS data processing model we do not foresee a real time online data processing. The processing farm is foreseen to be located at CERN. Experiment dedicated CPU farms like the one described above are already existing at CERN and are integrated in the CERN computing environment in the framework of the centrally operated SHIFT service, currently operating about 120 CPUs of different types. To minimise the network load at the first data processing stage a COMPASS dedicated tape server is envisaged. The possibility to integrate the COMPASS offline processing farm into the centrally operated computing environment is considered to be favourable over an offline farm operated by the experiment, especially in connection with the possibility of central data recording. A centrally operated cluster would also limit the load on the CERN network at later stages of the COMPASS data analysis.