By: Christoph Glingener

With each passing day, figuring out how to provision enough bandwidth for resource-intensive applications that connect to data centers and cloud computing becomes a more pressing, complex and costly question. The long-term solution lies in a more dynamic and programmable network infrastructure. Software defined networking (SDN) is a promising approach for achieving the cost-effective, end-to-end infrastructure flexibility that both network operators and their users seek.

The basic premise of SDN is to separate the network’s control and data planes, allowing the latter to focus on moving traffic as instructed by a logically centralized controller. For applications inside data centers, OpenFlow has emerged as one popular, simple and lightweight protocol that offers virtualization and control of a network environment through secure and standardized interfaces. Via an external controller, OpenFlow is used to specify flow-forwarding behavior across an enterprise’s infrastructure of electrical packet switches, thus streamlining and automating data-center operations.

Such implementations have delivered powerful benefits and revealed additional opportunities for even greater efficiencies, one of which involves the further virtualization of resources. Server and storage resources have already been virtualized; transport resources are next. Conventionally, data-center resources function in isolation from the operation of the transport network, but enabling the two to work together for optimal efficiency would help network operators and data centers avoid overprovisioning for peak load (plus the high interconnection costs and poor network utilization that can result) while also positioning the network to accommodate unforeseen bandwidth surges.

Envisioning greater potential for SDN

Imagine an online retailer during a busy holiday shopping season. Its network, computing and switching resources have been hobbled by the spike in transaction processing, so the retailer seeks to borrow from the infrastructure of an external data center.

As it stands today, if terabytes of data need to be exchanged and the static fiber connections are insufficient to support a bulk data transfer, a network manager or system engineer must manually divert load by shifting machines and executing a series of replication commands. This process, however, is time intensive. How much more beneficial would such a strategy be if SDN could instead be extended across the optical-transport network connection, allowing the fiber-optic pipes to be expanded on the fly as necessitated by application requirements?

SDN facilitated across network layers in this way promises truly elastic network operations and sets up the potential of virtualizing the entire network infrastructure, with a network manager seamlessly controlling all network, computing and switching resources and services via the same interface. Optical-transport capacity could be flexibly shared, optimizing resource efficiency and slashing interconnection costs, and programmable network control, faster adaptation to virtual machine (VM) mobility and simplified scaling would all be enabled.