Tradeoff study among traffic egression schemes and member allocation optimization for link aggregation groups in integrated switching systems

An integrated switching system (ISS) is an integration of individual switching devices with newly developed control and management protocols, in an integration topology that interconnects the switching devices. It scales the capacity of individual switching devices while functions still as a single switching facility to the outside network. In an ISS, a link aggregation group (LAG) has flexibility to allocate its member links across different switching devices, often as an attempt to spread risk of failure to individual switching devices. Due to existence of the integration topology, traffic admitted into the ISS and destined to egress out of the LAG interface may need to travel a number of hops before leaving the ISS. This lays a bandwidth burden on the integration links of the ISS. Local biasing and designated forwarding have been proposed as LAG egression options to relieve the bandwidth pressure on integration links. They bias traffic egression to be from the neighborhood of the switching device where the traffic ingresses, if the traffic is destined to the LAG interface and a LAG member link is present in the neighborhood. Those enhanced LAG egression schemes are in contrast to the regular LAG egression scheme, in which the traffic will be split and evenly distributed across all LAG members, regardless of their topological distance to the switching device where it ingresses. Although the enhanced egression schemes help to reduce the bandwidth demand on the integration links, there comes a price that load balance across LAG members may be sacrificed and eventually stable LAG capacity could be compromised. In this paper, we study such performance tradeoff and investigate impact factors on LAG performance. In the end, we formulate an optimization problem that optimizes LAG member allocation via pursuing the best tradeoff between integration bandwidth utilization and stable LAG capacity. The solution can be treated as a guideline to deploy LAGs in an ISS. The results show, with optimized LAG member allocation, the potential of integration bandwidth saving and stable LAG capacity maintenance can be maximally explored.