At last week’s Linley Tech Processor Conference in San Jose, there was a very interesting discussion about Software-Defined Networking.
In a keynote address, Christos Kolias, from the Strategy and Development Group at Orange’s Silicon Valley subsidiary, described the benefits of Software-Defined Networking (SDN), its impact on multicore processor platforms and some specific use cases.
As Mr. Kolias explained, existing broadband networks have become rigid, static and inefficient. Over many years, they have evolved to include a wide range of overlays (VPN, MPLS, VoIP etc) and a diverse range of discrete appliances (load-balancers, firewalls, monitoring devices etc). At the same time, routers are overloaded with many features that are used infrequently if at all, while many routing protocols are old and not stable. Finally, system interfaces tend to be closed and proprietary.
Mr. Kolias outlined some of the key challenges for future networks (such as video, virtualization, multi-tenancy and security) and stressed the importance for operators of being able to define their own routing and security policies. He pointed to trends in openness (open compute, OpenFlow), hardware commoditization and unified access. The inescapable conclusion is that today’s networks are too difficult to build operate and manage, given these emerging trends.
The key principle of SDN is that users can define traffic flows and decide what paths they take in the network. This concept brings a number of important advantages such as: the remote control of network hardware by software in a dynamic, programmatic fashion; the separation of control and data; the use of standard, open hardware interfaces; a full network virtualization plane. Operators benefit from the ability to run new services and applications, from added flexibility and control, and from opportunities for advanced optimization and customization. The concept of a network-wide OS is also supported.
The SDN concept is a good fit with multicore processor platforms. Mr. Kolias discussed how embedded multicore processors deliver quicker time-to-market than ASICs, provide transparent acceleration of networking functions through on-chip offload engines and lead to simple, open system architectures.
What’s your opinion on SDN? What are the key challenges that need to be addressed to ensure a viable transition from legacy network architectures to an SDN model? What are the implications for suppliers of processors, OSs and networking equipment? Is it reasonable to expect a shift to SDN over the next few years?