Software-Driven Wide Area Network (SWAN) is a centralized system designed to manage and control network infrastructure to enhance reliability and efficiency. Born from the need to address challenges faced by Microsoft Azure’s backbone network, SWAN aims to optimize traffic flow to meet varying demands of different applications. The idea for SWAN originated in 2012 and has since evolved into a fundamental system for Microsoft Azure.
Traditional backbone architectures were found inadequate to cope with the increasing demand for bandwidth due to Azure’s rapid growth. Hence, the concept of logically centralized control for both applications and the network was explored. SWAN’s development involved tackling algorithmic and prototyping challenges in a fail-fast approach, focusing on issues like fair resource allocation, memory limitations, and congestion management.
Prototyping revealed challenges with existing OpenFlow switches, leading to collaboration with Arista to develop DirectFlow for SWAN. This partnership enabled SWAN to support high-speed interconnects and SDN infrastructure for customers. The results of SWAN’s development were shared at the ACM SIGCOMM 2013 conference alongside Google’s similar system, B4, showcasing the potential of SDN in traffic engineering.
SWAN has continued to evolve, addressing additional challenges like equitable traffic distribution and proactive failure defense. Deployed in production on the Microsoft WAN, SWAN has significantly enhanced efficiency and flexibility. While some unforeseen issues arose, such as reliability concerns with centralized control during controller unavailability, new mechanisms were developed to mitigate such risks.
Looking ahead, Microsoft is expanding SWAN to other WAN connections, aiming to achieve greater efficiency and flexibility. SWAN now handles over 90% of the traffic in and out of Microsoft data centers, covering a vast network across all Azure regions. This unification is expected to unlock new levels of optimization, with Microsoft researchers focusing on future technical advancements in the field.
Article Source
https://www.microsoft.com/en-us/research/blog/born-in-the-research-lab-a-decade-ago-swan-continues-to-accelerate-networking-in-the-microsoft-cloud/
