The filter is known as the "trigger". The trigger has several levels, so that the data rate is reduced by an order of magnitude or more at each level. The lowest levels are done in hardware for speed. The upper levels are in software for flexibility.<p>Designing the trigger is extremely complex, since the various detectors within each experiment have greatly varying response time. Only data from fast detectors is available for the low level trigger.<p>In addition, the speed of light is a real barrier for the lower levels of the trigger; by the time the debris from a collisions reach the outer reaches of the experiment (this is usually where the muon chamber is), there have already been an additional collision at the center.<p>(The speed of light is about 1 ft/nanosecond, the radius of the muon chamber in CMS is about 25 ft, and the time between bunch crossings is about 25 nanoseconds.)<p>The design of the trigger is a very important and often contentious process. A bad trigger will throw out important physics events, and trade-offs can favor one physics search (e.g. the Higgs) over another (e.g. supersymmetry).