Simple Failure Resilient Load Balancing

To enable reliable data delivery and balance load in the presence of failures, we propose a new mechanism that combines path protection and traffic engineering. The key benefit of our solution is its simplicity, allowing for fast recovery while imposing minimal requirements on the routers. To provide resilience against every failure scenario from a known set, we advocate using a fixed set of parallel end-to-end paths for each traffic demand. Upon detecting a path failure, the ingress router uses a local rule to rebalance the outgoing traffic on the remaining available paths. We describe several candidate rebalancing algorithms, and analyze their performance. Although calculating the optimal set of paths and the path-splitting parameters for each router is NP-hard, our extensive simulations on a tier-1 IP backbone demonstrate that our easy-to-calculate heuristic suffices to achieve nearly optimal load balancing. We believe that a simple-to-implement solution with a fast recovery time, such as ours, will appeal to Internet Service Providers as well as the operators of data centers and enterprise networks.