IT buyers are under unprecedented pressure to evaluate the impact of software-defined networking (SDN), the application ecosystem, and all of the implementation challenges. But software-only SDN, optimized to run on specific off-the-shelf hardware models (or software generic enough to be server-agnostic), is handicapped performance-wise and probably not much cheaper in the long run.
While the cost benefits for an SDN platform are still speculative, the business case to support SDN does not seem to be in lowering capex or opex. Rather, network agility and the the promise for greater levels of control, innovation, and flexibility in the network are where the advantages lie. SDN has the greatest potential to boost network utilization and resiliency in applications where a large number of servers and WAN network connectivity mesh.
A true SDN service tool is real architecture that covers the entire network landscape and extends through all the service elements to the management layer. Management automation with secure, controllable properties is the only path to optimal operating costs.
To stand out, SDN developers need to offer the next big thing in performance — including hosts and caches and software features. Their products need to carry more traffic; reduce the time it takes to facilitate the network and security services around a new workload or virtual machine; segment the network to provide highly secure, multi-tenant services (either public cloud or enterprise); and so forth.
A software-only solution will make little progress on those fronts. Creating virtual functions running on white-box hardware creates complex operation challenges that don’t necessarily boost performance.
One concern regarding custom hardware is that the SDN market is young — the market, standards and technology are evolving rapidly. However, the risk is that the white-box architecture will not be able to implement the next generation of features at the next generation of speeds. Existing white-box implementations are limited in rules and bandwidth, while NPU-based architectures are on an ever-expanding NPU roadmap. By sheer nature of using a packet-oriented NPU over a more generalized architecture, application developers can add protocol support in a few days, all while maintaining the throughput of layer protocols.
The next generation of NPUs will have exponentially more bandwidth and processing power specifically focused on packet-oriented applications. Combined with newer and larger TCAMs, the technology roadmap contains everything needed to create a future-proof product line.
In short, for best-effort packet delivery, access control, tunneling, or data encryption, a critical component for SDN performance is the hardware platform itself. An SDN-supported node, in particular, plays a crucial role, influencing the possibility to realize SDN-specific features for given traffic.
Hence, the message to software-focused developers: if you don’t use next-gen silicon technologies to create a competitive advantage, someone else will.
Whether it is an OpenFlow controller, SDN switch, or router, a solid hardware design foundation considers switch technology, processor requirements, I/O options, physical size, cooling requirements, custom features, private labeling, time to market, and regulatory certifications, along with forward-looking aspects such as next generation offering, model variations and any other features that offer a competitive advantage.
Great article Matthew. I am currently deployed at a whitebox switch vendor. I’m curious if they have thought about th perspective this article shares.