Data Center Challenges: Building Networks for Agility

The data centers used to create cloud services represent a significant investment in capital outlay and ongoing costs. Using a "follow the money" approach, I'll outline some of the major challenges facing the creation of cost-efficient cloud service data centers. I'll then examine the problem of the intra-data center network in depth and show how services hosted in today's data centers suffer from internal fragmentation of resources, rigidity, and bandwidth constraints imposed by the architecture of that network. I'll then describe Virtual Layer 2 (VL2), our approach to a new architecture for the network, and explain how it can be built today out of readily available components. A prototype of this architecture demonstrates the approach actually works and provides excellent performance, shuffling 2.7TB of data among 75 servers in 395 seconds – 94% of the best possible utilization. Our work supports several conclusions: (1) randomization works well as a method for coping with volatility and traffic engineering in data centers; (2) using many low-cost switches in topologies that create extensive path diversity is a simple and effective method for coping with failures; (3) existing layer-3 protocols can be leveraged to build cheap and resilient data center networks; and (4) simple software modifications in the host networking stack can provide the network virtualization that data center tenants desire while eliminating the complexity of network configuration — this software also provides hooks that can be used to work around limitations of the cheap switches and add other functionality to the network.
David A. Maltz is in the Networking Research Group of Microsoft Research where he designs new architectures for data center and enterprise networks and investigates techniques for managing IT infrastructure. He was a founder of startup companies in network traffic management and wireless networking, leading engineering efforts at each.