The invention of the multicore processor was a great achievement with significant influences on the throughput, performance, and speed of computing. It has been the origin of two major investigation streams: how to exploit its untethered power and how to imply the appropriate concurrency in the codes perfectly. The first subject is the main aim of this paper.

Devising an efficient policy to assign the packets to the cores is the objective of the paper. The packet scheduler is a key component in the multicore network processor architecture. Traditionally, packet-level and flow-level load balancing schemes are deployed to assign the packets to the cores. Dispersal of the packets’ movement among the cores is one of the properties of the packet-level load balancing method. To mitigate this issue, the authors discuss a flow-level load balancing method that operates based on the hash function to assign the flow of packets from an application to a target core. Although this policy affords the caching and ordering advantages, its efficiency depends on the quality of the hash function. The monotonic nature of hash scheduling is not like the asymmetric dynamism of packets generated by the applications. Even with the best distribution of flows to cores, skewed load and flow bundle would be the next issue. The paper proposes mechanisms not only to keep the advantages of caching and ordering, but also to balance the loads to different cores with minimum performance side effects.

Migration of packets from the overloaded nodes to the underutilized ones is the favorite policy to implement load equilibrium. The authors provide a two-level annex-cache hardware mechanism, an aggressive flow detector (AFD), which not only identifies the high aggressive flows in the network, but tries to balance their load symmetrically with less overhead by using an appropriate migration policy. AFD constitutes a small fully associative cache, aggressive flow cache (AFC), and an N-way associative annex cache.

Multiservice core allocation was discussed perfectly as well, covering two aspects: how the core is assigned to the service and how the packets are conducted to their favorite service cores.

The achieved results and performance of the proposed architecture are evaluated by simulation, and the results indicate the improvement of throughput and performance by utilizing a locality-aware packet scheduling (LAPS) policy.

A comprehensive, implementable proposal has been propounded. It embraces plenty of useful discussions about the hardware techniques and an analogy from networking science. It is definitely is a fruitful reference for those in the multicore research community.