In the prior post, I talked about the cloud-scale SDN market, in this post we’ll take a look at the second market segment marching down the path (or at least checking the route on Google Maps) towards SDN – the Service Provider market. Service Provider solutions are themselves further delineated by their target application: Core Networks, Metro Ethernet, Wired Access and Cellular Access. Additionally, Service Providers delivery data center solutions (hosted, co-located and cloud), however these deployments will ultimately follow the trends set by the cloud-scale providers, albeit likely requiring the completion of work started by groups like OpenStack and the Open Compute Project. The Service Providers sit squarely in philosophical camp of The Architects. They are looking for toolkits which enable them to drive down their costs basis, increase their agility and compete effectively against the cloud-scale providers.
One approach to the Service Provider SDN solution would be to treat the entire network as a single domain, however this methodology is critically flawed. Scalability of the flow tables resident in the network equipment and of the centralized flow-setup engines precludes such an approach. Instead, the problem space will need to be broken down into tiers, with well-defined abstractions which will provide the necessary scalability in the core. Core backbone network solutions will need to be geared towards macro-flow traffic engineering and traffic re-routing on link failure. It would seem that sufficient scale can only be achieved by aggregating traffic into classes and dealing with network end and mid-points (not customer or device level granularity).
Metro Ethernet solutions will likely be focused on the same target applications, though with a requirement for finer grain visibility and control (enterprise customer and class of traffic). This also the likely spot for NFV service chaining to be an integral part of the overall solution. Wired and Cellular access networks are similar, but require an even more granularity (individuals vs corporations). This increase in granularity flows through to the back-end services being provided and increases the complexity and scalability requirements of the service chaining solution.
Broadly speaking, SDN has to provide two compelling arguments to successfully transition a market away from the status quo: a significant reduction in the capital required to implement the solution as well as transformation in the operational administration of the solution. This is what drove the cloud-scale datacenter solutions towards success, and we have to identify how these attributes can be unlocked in the other networking markets. In the context of SDN, the Service Provider market presents several unique challenges. Traditionally, the majority of the network equipment in the Service Provider networks has been highly customized and in those areas where it has become more standardized (Metro Ethernet) the chipsets typically leveraged do not currently have sufficient scale to address the granularity required of the solution. Secondly, the applications currently deployed by the service providers to address their operation requirements are highly customized and exceed the scale of existing SDN controllers and hardware.
I see the two areas most amenable to adopting SDN technologies as the WAN Core networks and the Service Provider datacenter solutions. The WAN core will require new software stacks to be deployed on the network elements (if not new hardware as) and controllers with the scale and resiliency not yet seen (although ON.Lab is working on it). The datacenter implementations can mirror the hardware choices already made by the cloud-scale providers and being standardized in the OCP Networking working group, but also require the rest of the datacenter infrastructure to be instrumented and orchestrated. OCP and OpenStack are leading the way towards the ecosystems required for this to become reality.
What do you think? What are the key hurdles to be overcome for the Service Providers?
Leave a Reply