Balanced locality-aware packet schedule algrorithm on multi-core network processor

Previous work has shown that processor affinity is one effective way to improve the performance of SMP systems. A detailed analysis of cache characteristics on network processor is carried out. The result shows that network packet processing can also use cache affinity to gain performance improvements by reduce the miss rate of instruction cache and data cache. A schedule algorithm for multi-core network processor taking both load balancing and packet affinity into account is proposed, called BLA. BLA try to schedule packets from the same flow to the same core while keeping work load balanced between cores. The result shows that, BLA can reduce the number of missed cache access by 13%, while the load difference between cores is controlled within 2%.