In my last blog I pointed out that we really needed to understand what the end-game network would look like to effectively drive SDN and NFV. Today I’d like to look at how the possible end-games relate to the vendors in the game. Who might win, or even drive, a specific SDN/NFV future?
There seem to be three possible SDN scenarios that lead to my utopian SDN model. They are the data-center application and service path, the up from the depths path, and the magnetic service path.
In the data center application and service path, SDN develops as an application and tenant segregation concept within a cloud data center. This requires a fairly sophisticated virtual networking approach, much like Amazon and Google (whose Andromeda approach is fairly public) have developed. Network virtualization in a data center can be leveraged to extend beyond the data center, and that would be the way this specific path could be realized.
The critical step in this path to SDN is that extension. Even now, we have vendors like Nuage who have an overlay model of SDN that can easily be extended to the edge, and so Alcatel-Lucent is obviously a competitor who could make hay with this opportunity. Brocade, who has a stronger data center positioning for SDN, could leverage its Vyatta offering with edge-hosted virtual routing and tunnels to build a similar uniform end-to-end virtual network model. Juniper, with Contrail, could pair SDN and fabric switching to at least create the data center part, but it’s not clear how easily they could take the next step.
My view is that this approach to SDN success is the most “open” meaning that virtually any data center SDN strategy could be extended to complete virtual networking through either product extensions or partnerships with any number of players. Open-source tools like OVS could also play a role, as could an open-source router. The very breadth of possibilities may render this path, ironically, less attractive. That’s because anyone who takes the trouble to educate the market in a broad SDN transformation via the data center will carry a lot of competitors along too. Who wants to do that?
The second possible path for SDN is the “up-from-the-depths” path, which means starting with agile optics and optical grooming, extending that in some way to provide better granularity of tunnel management at the electrical layer (virtual L1), and then marrying these L1 partition-based networks with virtual switching and routing. This model is easily applicable to business data services and with some care to mobile and CDN applications. It’s a natural play for a network operator, in other words.
You need to be an optical player to win here, and in theory Alcatel-Lucent, Ciena, and Infinera would be obvious candidates. Vendors like Juniper and Cisco could also jump in because the critical extension in this model is that electrical L1 grooming, which even someone who isn’t really an optical giant could do. What you do need is to be able to add electrical L1 grooming to fat lambdas.
If I had to bet on who’d take a shot at this path it would be one of the pure-play optical types like Ciena and Infinera. This path would undermine carrier router and switch deployment in favor of per-customer or per-service virtual switching and routing. That would make it less attractive to Alcatel-Lucent, Cisco, or Juniper. The wild card on this path would be NFV, of which more will be said below.
The final SDN path is the notion of a “magnet service”, something new and profitable. One could argue that the only qualifying candidate here is the Internet of Things, but IoT is so mired in useless hype that I’m not sure how easily a logical approach could be promoted. In any event, it’s very likely that the only people capable of using IoT to pull through massive SDN would be a service provider. Most of them seem caught in the trap of thinking that every controller and sensor in the future will have a dedicated 5G connection that they can sell a user for twenty bucks a month or so. Count the sensors and controllers in an average home and you can see how far that would get. I think that while almost any SDN player could claim to support IoT as an SDN driver, I’m doubtful that it can drive SDN by itself. Again, more below.
On the NFV side, the possible paths to the future are the operations umbrella path, the getting-cloudier path, and the old favorite magnetic service path.
The best way to make NFV deploy quickly and widely would be to transform service and network operations with model-based orchestration and automation. That’s been clear from the very first, but even vendors with good credentials were slow to push on this approach. Oracle came along early this year to take advantage of the void in operations positioning, and other vendors are now saying (and doing) more. The functional leader in this space is HP, then Overture Networks, then Alcatel-Lucent and Oracle. There are other players who appear to be working in this area, notably Ciena and Huawei, but I can’t yet say just where either really is on the topic of operations integration.
The reason strong operations integration would be so compelling is that both operations efficiency and service agility benefits—the primary drivers of NFV in operators’ view—depend on service/network management automation and coordination to reduce human interaction and speed the handling of service and network events. If this capability is provided at a high level (the OSS/BSS and NFV MANO levels) then making the business case for NFV on a per-service, per-PoC, basis would be much easier. The classic example is “virtual CPE” hosted on an agile premises device. With an operations umbrella on top, this model is easily validated based on first-cost controls and agility alone. Without that umbrella you can’t even confidently assert that it saves anything at all.
I think we’re going to get some action in this space, but I’m concerned about the timing. The only way to socialize the operations umbrella strategy is through the CIO, one of the executives that hasn’t been much engaged in NFV trials. Most vendors aren’t calling on the CIO regarding NFV, and they don’t know whether to position operations orchestration as an OSS/BSS transformation or ride on NFV’s coat-tails. If there’s not some quick action this fall, we may go into 2016 budgeting with no clear path to an operations-driven field trial.
The cloud-driven approach to NFV deployment says that nothing much in the network (meaning connection) services area is going to move the ball fast and far enough. Operators still see cloud computing as a major revenue opportunity, and if they pursue it they will be building cloud data centers that look structurally like NFV infrastructure and that present deployment and operations challenges similar to that posed by VNFs. Operating the cloud efficiently at scale is arguably an NFV-like problem.
HP and Brocade could both drive this evolution in theory. HP has both SDN and NFV assets, and both would be valuable in a cloud-path NFV success story. However, vendors have been generally soft on making the connection between NFV and the cloud. This may be a situation where a player like Intel, who offers foundation tools for the cloud and for NFVI, might want to step up and make the point to insure this path gets a fair hearing.
The final topic for NFV is the old familiar “magnet service” and as is the case with SDN, the most credible (perhaps the only credible) magnet is IoT. With NFV, there may be more magnetism available, because NFV could deploy the query and collection functions needed to gather information from sensors with arbitrary connection options (often through intermediary gateways) and send information to controllers similarly connected. There is a network mission here, but in many ways it’s a mission that’s very cloud-like or service-like, which is what makes it an ideal NFV story.
Alcatel-Lucent, HP, and Oracle are all developing (and in all cases, have developed at least in a preliminary sense) an IoT architecture that would at least involve NFV. None of them are complete at this point, or at least not completely disclosed. I think this is largely because we’ve still not accepted a truly comprehensive and sensible model for IoT overall. It’s going to take a bold vendor to step up and blow against the prevailing media winds, but that’s an essential step if we ever hope to realize IoT potential. And if a vendor does prevail in promoting a sane model, they’ll have a chance to establish themselves both as an SDN and NFV kingpin.
It seems clear that IoT is the wild-card stimulus for both SDN and NFV, in part because it is a stimulus for both. No matter how many current hopefuls we have in the space, the fact is that we have no convincing positioning or fully articulated strategy (HP is the closest) relating IoT to either SDN or NFV, so anyone can claim the prize.
We have at least competing claims in all the driver areas, in fact, and that means that we’re entering the critical year for both SDN and NFV without a clear winner, and perhaps given the multiple possible paths to success, without a clear leader. It should make for an interesting fall and Q1, but I think things will shake out sharply in the second half of next year.