Software-defined traffic load balancing for cost-effective data center interconnection service

Abstract

For interconnection between geographically-separated data centers, network carriers typically implement multiple optical paths in a wide-area network. For example, when transmission wavelengths have 100 Gb/s granularity, three 100 Gb/s wavelength paths are provisioned to satisfy a customer demand of 300 Gb/s. Over the multiple provisioned paths, interconnection traffic is typically distributed using per-flow hashing, which results in an uneven distribution of traffic caused by hash collisions. For a relatively-few number of high-bandwidth traffic flows (> 1 Gb/s) between data center locations, per-flow hashing can perform poorly in terms of bandwidth utilization and availability. We propose new software-defined traffic load balancer, SD-TLB, that performs measurement-based flow distribution over multiple optical paths, with an implicit impairment detection method using per-port statistics on available paths and a flow redistributor that is immediately adjusted to the current network state. While our approach does not provide the same level of protection as 1+1 optical protection, it can provide the necessary redundancy for data center inter-connection at a lower cost. We experimentally implement the SD-TLB using ASIC-based switches and open virtual switches interconnected by wavelength-division multiplexed transport network test-bed. The experimental results show that SD-TLB outperforms today's hashing-based alternatives in balancing, throughput, and restoration in the presence of outages and impairments and as a result achieves improved cost-efficiency.