Network Processor for
Tomas Henriksson, Ulf Nordqvist, Dake Liu · 2003
Computer networks use higher and higher data rates. This leads to bottlenecks in the protocol processing in the terminals. The protocol processing can be offloaded to a special unit for a more efficient implementation. The algorithms used for protocol processing have inherent parallelism, which is best exploited at thread level by a multiprocessor architecture. Key is to partition the tasks and allocate them to the processors in the system. In-line processing reduces the memory bandwidth requirement and the processing latency. We have developed a partition scheme that works well with in-line processing in a dual-processor system.