Performance Optimization of Scalable Coherent Interface (SCI) Systems

Experiments in modern physics deliver huge amounts of data with very high requirements of data acquisition, buffering, sorting and storing. Within these requirements data has to be delivered to a processor farm for the calculation of trigger data. These processors have to be interconnected with a scalable network which guarantees the necessary throughput even in heavily loaded systems. This project deals with an adaptive control mechanism to reduce the retry traffic in heavily loaded systems and therefore improves the overall performance of SCI-Systems under overload conditions of up to 50%. This and the suggested combination of SCI-switches to complex systems guarantees the special requirements of data-exchange within computer farms.