Datacenter networks need to handle traffic loads of varying characteristics and quality-of-service requirements. Notably, traditional workloads such as Web search and data mining have very different Quality of Service (QoS) requirements and thus need to be serviced accordingly. Network-level load balancing algorithms have been proposed to help provide flows with adequate QoS by assigning traffic flows to different network paths to avoid congestion and improve overall network utilization. As datacenters scale up and get distributed geographically around the world, there is a growing trend of wide-area network (WAN) datacenter traffic, which typically consists of data-heavy tasks, competing with intra-datacenter (DC) traffic. A noteworthy challenge raised by the interaction between DC and WAN traffic is the differences in link utilizations and round-trip times. As such, it is imperative to design load balancing algorithms that meet the QoS requirements for both traffic types optimally. To the best of our knowledge, existing datacenter load balancers have not addressed the coexistence and interaction of WAN and DC traffic. To highlight this gap, in this paper, we conduct a comparative performance study of state-of-the-art datacenter load balancers considering  different datacenter topologies and different workloads. We study existing load balancers when subject to both DC as well as WAN workloads. Additionally, we provide recommendations on how to set load balancer parameters and share our experience in engineering experimental workloads. We conclude with a discussion on how adaptive load balancers can address the challenges posed by emerging and future datacenter topologies and workloads.

Event Host: Lakshmi Krishnaswamy, Ph.D. Student, Computer Science & Engineering

Advisor: Dr. Katia Obraczka