Massively parallel priority queue for high-speed switches and routers

The need for better quality of service is growing as new quality sensitive services are becoming more and more important in data networks. The key element in providing quality-of-service or grade-of-service in packet networks is deciding when or in what order the incoming packets should depart the system. In other words the departing packets have to be somehow scheduled. Implementing this scheduling becomes harder and harder as line-speeds and number items to schedule increases. One way to implement the scheduling function is to use a priority queue that can be realised by using simple and scalable heap data structure. The problem is that traditional sequential heap cannot be made fast enough. In this paper a massively parallel system based on set of heaps is presented. It will be shown that such system can be scaled to support very large priority queues at STM-256 speeds and beyond.