Practical multipath load balancing with QoS

Abstract

The Internet is based on a single-path communications model. This model imposes significant constraints on the ability of the Internet to satisfy the quality-of-service requirements of network applications, and results in significant inefficiencies in the use of network resources that are manifested as congestion. The result has been the need to over-provision Internet-based systems to meet the basic needs of modern communications. With the adoption of the Internet as the converged communication infrastructure for the 21st century, this is clearly not an acceptable long-term solution. One approach that has been identified to address these limitations is to enhance the Internet routing architecture to support multiple paths between a given source and destination. Significant research has been done into multi-path solutions for QoS and congestion, however a comprehensive solution for both QoS and congestion that is compatible with the Internet's datagram, hop-by-hop model of communication is still elusive. This paper reviews a solution presented in previous work, called Dominant Set Multipath Routing (DSMR), that addresses these requirements. The DSMR algorithm computes the best set of routes between each source and destination that provides the full range of performance available from the network. This set is used to route flows over paths that both meet the QoS requirements of the flow and minimize congestion in the network. Simulations are then presented which show the effectiveness of DSMR to provide 3 to 11 times the capacity of single-path routing while meeting QoS requirements and minimizing congestion.