Raw Data - Brute Force

The multilevel trigger approach aims at reducing the number of events to be accepted in several stages. The first level of the CMS detector will be implemented entirely in hardware. This logical device receives data from a subset of detector elements to come to its decision. It reduces the event rate from $10^9$ Hz to $10^5$ Hz. This trigger level provides information about its results for further event building. Examples can be seen in table 5. This information is used as a first hint for dynamically selecting higher levels of algorithms for event processing. Another task is to synchronise the data of all subdetectors[NTWS97]. Due to the latency of the detector elements and readout delay it is necessary to globally tag the data fragments. Only then can they later be associated with the same collision event. The system is custom built and fully pipelined. Processing even a simple algorithm takes already several bunch crossings time. Therefore incoming information is buffered in a pipeline and the pipeline entries are processed in parallel with a maximum latency $\le 3\mu$sec. As the number of processing boards is in the order of 10000, cost is a major issue. To keep the pipeline that has to buffer events small, the trigger has to work at high speed (160 MHz or more). Not all subdetector information can be used for first event rate reduction, but only calorimeter and muon system. The trigger is programmable to a certain extent, i.e. simple functions can be specified and thresholds can be modified. The overall amount of summary data that is provided is a chunk of 10 KB per event (see table 4). The global trigger information must be read for every event before data from other subdetectors can be examined. The sizes of the data chunks that are eventually used to describe the event are outlined in table 4). The output rate of the level 1 global trigger is 100 kHz, which results in a total event size of 1 MB at $1 \times 10^{11}$ GB/s (= 100 GB/s).