Adaptive Load Sharing for Network Processors

Abstract

A novel scheme for processing packets in a router is presented, which provides for load sharing among multiple network processors distributed within the router. It is complemented by a feedback control mechanism designed to prevent processor overload. Incoming traffic is scheduled to multiple processors based on a deterministic mapping. The mapping formula is derived from the robust hash routing (also known as the highest random weight-HRW) scheme, introduced in K.W. Ross, IEEE Network, 11(6), 1997, and D.G. Thaler et al., IEEE Trans. Networking, 6(1), 1998. No state information on individual flow mapping needs to be stored, but for each packet, a mapping function is computed over an identifier vector, a predefined set of fields in the packet. An adaptive extension to the HRW scheme is provided in order to cope with biased traffic patterns. We prove that our adaptation possesses the minimal disruption property with respect to the mapping and exploit that property in order to minimize the probability of flow reordering. Simulation results indicate that the scheme achieves significant improvements in processor utilization. A higher number of router interfaces can thus be supported with the same amount of processing power.

An efficient packet scheduler for modern network processors: guarantee load balancing and packet ordering

“…As mentioned in [6], a good measure can be the number of memory accesses or the number of processing cycles an NP has performed during the time interval ) ,…”

“…Obviously, a tradeoff should be made between load balancing and packet reordering by using different values of threshold pair. This tradeoff will be discussed in the following sub-Section D. Finally, we compare our F2CL-based scheduling scheme with other three popular packet scheduler, HRW [6], IBM [3] and SHI [7]. Our F2CL is configured with m=8, =95%.…”

“…Our hashing method follows its analysis results. IBM Research [3] introduces three flow assignment techniques to achieve load-balancing among multiple NPs, which are timestamp, flow [6] proposes a scheduling algorithm for parallel IP packet forwarding, based on robust hash routing (also known as the Highest Random Weight, HRW). Further considering the adaptation disruption of load scheduler, W. Shi et al [7] divide Internet flows into high-rate and low-rate flows, and only schedule packets in the high-rate flows, while lower-rate flows are mapped to FEs by the hash splitter.…”

A Practical Packet Reordering Mechanism with Flow Granularity for Parallelism Exploiting in Network Processors

“…Although flows can be reassigned from overloaded PEs to idle ones to retrieve fair traffic distribution, PD problem is inevitable. Kencl et al in [5] present a robust hash algorithm to minimize the number of reassigned flows, but it can't eliminate PD completely and the computation cost is too large to implement the algorithm in a real system. While in the system with UPSL, the above problems are not so frequent.…”

“…Since packets can be processed without flow limitation, all PEs can be made full use of if packets are dispatched fairly, and throughput under bursty traffic won't be decreased greatly. Even in a system with SPSL described in [5], UPSL can still be used to ensure packet order of reassigned flows. However, in the area of NP, the research literature contains surprisingly few papers presenting how to implement such a system and what the critical issues are especially with flow granularity; neither has industry paid enough attention to this problem.…”