Evaluating native load distribution of ARP-path bridging protocol in mesh and data center

Abstract

ARP-Path is a simple, low latency, shortest pathbridging protocol for campus, enterprise and data centernetworks. We recently found that this protocol nativelydistributes the traffic load in networks having redundant paths ofsimilar characteristics. The reason is that every new pathbetween hosts is selected on-demand in a race among ARPRequest packet replicas over all available paths: the first arrivingreplica gets its path selected on the fly. This means a continuousadaptation of new paths to variations on the load at links andbridges. To show this unique load distribution capability andpath diversity property we use a number of simulations forcomplex scenarios, including two different simulators: one flowbasedand one packet-based, and two basic topologies: datacenter and a regular mesh. We also verify this behavior on realhardware on a network of nine ARP-Path NetFPGA switches.The conclusion is that the ARP-Path protocol efficientlydistributes traffic via alternative paths at all load levels, providedthat multiple paths of similar propagation delays are available