Wall Street has generally focused its research and analysis on how SDN will impact the technology sector. I have also expressed my views on this topic in prior blog posts and have generally taken the view that Layer 4-7 appliance companies may be most at risk as such appliances will be replaced by software applications, merchant silicon semiconductor companies may be poised to benefit as replacement cycles compress for networking equipment once the control plane is detached from switches, and the jury was still out on traditional switching/routing companies depending on how these companies maintain some level of software differentiation over emerging “white box” networking suppliers.
Last week I attended the OFC/NFOEC optical conference and walked away with some additional elements of my evolving SDN investment thesis. In particular, while traditional telecom operators should benefit from the potential benefits of deploying SDN in their network, they may be also be at risk as SDN will move the value away from the physical network to the application layer where differentiation will be determined by using software for service creation. While traditional telecom operators are clamoring for SDN as a way to reduce vendor lock-in, lower network cost and enhance service creation, I am not yet sure how well they will compete against software-centric rooted large data center operators like Google in cloud computing services. Thus, as the value moves away from the network to the software layer, SDN may actually be a threat to traditional telecom operators. Companies that can help traditional telecom operators through this transition to allow them to better compete vs. software-centric data center operators, will ultimately derive significant value in the financial markets in my view.
While OFC/NFOEC is supposed to be a conference specializing primarily on optical communications, SDN permeated several of the presentations and seminars. What I found interesting in several of these presentations was the contrast of how large data center operators like Google and Facebook talked about their specific traffic patterns and resulting approach to building out their data centers and network and how they plan on using SDN in this regard vs. how traditional telecom operators discussed the same topic. The following table shows some general initial takeaways I had from these presentations.
Software Centric Data Center Operator
Traditional Telecom Operator
|Traffic||Mostly machine to machine||Mostly end user driven|
|Hardware||Disposable Asset||Long Term Asset|
|Software||Core competency||Bundled By Vendor|
|Network Protection||Algorithm Focused||Network Focused|
|Benefits of SDN||Service creation
Remove vendor lock-in
Let me reflect on a few points on the above table. Large data center operators like Google and Facebook are fundamentally software companies while traditional telecom operators are generally not. The ultimate virtualization of the network layer, which is a key objective of SDN, will make software more of a differentiator between data center operators than it is today and could further differentiate data center operators in business and cloud computing services vs. traditional telecom operators in the future.
For example, a large data center operator at the conference talked about how they replace their servers every 18 months in their data centers as it is more cost effective for them to purchase new servers than to run their data centers on older servers. Now I am not sure what the replacement cycle for servers are in data centers are large telecom operators, but the mindset of hardware being disposable is not typically embedded within the culture of traditional telecom operators.
Another example that resonated with me at the conference was how several telecom operators (and data center operators) talked about how optical transport cost is now about 80% of the network core capital spending costs vs. 20% for routers whereas several years ago the percentages were exactly the opposite. In addition, some of telecom operator presentations also talked about how network protection in the optical core sometimes equates up to 50% of the network cost. So, if routing is becoming a much lower relative cost in the core than optics, why are telecom operators putting so much focus in SDN presentations on vendor lock-in within Layer 3 of their network? Clearly all types of cost reduction should be pursued and attacking 20% of the cost is still important, but if separation of the control/data plane in Layer 3 is only going to address 20% of your cost, perhaps there should be more focus on industry standards for optical layer control protocol (e.g. extension of Openflow to the optical layer) and API software development that attacks network utilization and restoration.
So in summary, my main conclusions from the OFC/NFOEC conference in relation to the evolving SDN market are:
- While traditional telecom operators will benefit from SDN, they may also be at risk given a more software centric culture and pedigree at certain large data center operators. Companies that can help traditional telecom operators becoming more software savvy will likely become valuable companies.
- Optics is becoming a larger part of the network cost problem than routers for both data center and telecom operators. Hardware and software companies that attack and solve this problem will likely become valuable companies. Although funding for such hardware initiatives is not in vogue, hardware companies could include merchant silicon companies for coherent optical DSPs or companies that innovate on integration of optical components (e.g. silicon photonics, indium phosphide). Software companies could include companies that solve high costs associated with network utilization (given the very wide spread in network traffic between peak and average traffic loads) and network protection.
- While switching remains an important cost problem, it presents a much bigger problem within the data center in terms of network agility and an obstacle to service creation. Data center operators want switching solutions that scale horizontally with the control plane disaggregated.