The CDF experiment, known for its discovery of the top quark, deploys a three level event processing architecture[Pau99]. The first level retrieves data from the detector electronics and is fully implemented in hardware. It outputs to the second level which is realised using commercial processor boards, but custom-designed software which is specific to the subdetector data. This level serves events to the level three processing farm at a rate of 300 Hz to 1 kHz, depending on the chosen system configuration. The event size varies between 150 and 250 KBytes[TBD$^$97]. This level is connected to the level two data servers using an ATM switching network. Non-blocking message sending behaviour is achieved by applying traffic shaping at the data sources based on time division multiplexing. Although monitoring and control software is implemented using object-oriented technologies, such as Java and CORBA, all software associated with the data dissemination is done in pure C [Tet99]. Several libraries, such as FISION/VISION[Pan97] are deployed for less error prone access to VME hardware. In order to interface to the object-oriented software, the ILU[XER99] object request broker has been ported to the VxWorks real-time system. The CDF experiment has been designed at a time when the application of object technology was not yet suitable for real-time systems, because of performance problems and a lack of tools. The decision to introduce newer technology gradually came with the possibility of granting the detector a second run phase with increased beam luminosity and detector precision. Yet a complete object based solution was not adoptable, due to the enormous amount of work required to redesign the software of the complete data acquisition system.
