Sometimes really important things reach the news but in pieces, and that may be the case with two items in SDx Central. Huawei, says one article, is beefing up its open source credentials with new hires. Fujitsu, according to a second article, has announced a new SDN controller that uses two levels of abstraction. It’s not clear reading the pieces that these two items are related beyond an open-source reference, but I think they are, and that the relationship is a milestone in the evolution of both SDN and NFV.
Operator interest in open source is real, but there’s a big defensive component to their thinking. Virtually every operator, and every Tier One I’ve talked with, thinks that network equipment vendors are dragging their feet on the evolution to SDN and NFV to protect their own business models. Open source lets operators base their evolutionary strategies on software that (in theory) vendors don’t control. The reason for the qualifier is that vendors dominate open source as much as they dominate standards bodies, for the same reason. More bodies and money to spend. But it’s harder to drive collective development in your own selfish direction than it would be to drive your own development that way, so operators are hoping that open source will weaken vendor strangle-holds on progress.
Open source is a priority, but there are a lot of open-source projects out there. Huawei’s two hires, the topic of the first piece, are involved in OpenDaylight and OpenStack, which are surely open source but are more directly related to NFV when you take them together. Which, in timing terms, is what Huawei has done. If you were planning a big NFV push, you might have a specific interest in ODL and OpenStack.
But Huawei already has an NFV strategy. That’s where the second piece comes in. Fujitsu’s interest in a two-level abstraction for ODL, meaning that the northbound interface is abstracted (pretty much as usual) and the southbound (to the devices) is also abstracted. This second level of abstraction means that you can use a modeling language (YANG) to map devices to the southbound model, and that the ODL controller actually talks to the abstract southbound interface only—the actual device mappings are handled by the model.
Still looking for the tie-in? Well, Dave Lenrow, the guy Huawei got from HP, happens to be perhaps THE go-to guy for the hot topic of intent modeling, including as a northbound interface for ODL. Intent modeling seems the basis for the Fujitsu dual-abstraction model too. So one thing we seem to have here is a common thread of intent modeling, a linkage between the Huawei and Fujitsu stories that’s more complicated (and important) than the open-source connection.
For Huawei, intent modeling could be a critical addition to its NFV strategy for several reasons. First, every operator (including both AT&T and Verizon in the US) is increasingly aware that you can get to NFV operations and agility benefits only if you can bind both legacy and NFV elements in a common infrastructure pool. That’s a specific goal in a preso Verizon recently made; start your evolution by creating an operations framework that abstracts the infrastructure away.
For Fujitsu, the dual-level abstraction is critical if they’re to have either SDN or NFV success because they have to be able to work with other vendors’ gear to get their foot in the door. The dual-abstraction approach lets them address any kind of gear while at the same time keeping the core implementation of the SDN controller clean. It also facilitates a common management view from the devices upward, because the southbound model can handle the management abstraction too.
So for sure, abstraction is getting critical. I think it’s even more specific, as I’ve said, and that intent-model-based abstraction is getting critical. As I said above, and as I’ve said in prior blogs, you can’t achieve operations efficiency or service agility if all you can operationalize is that tiny NFV enclave you’ve started with. The intent-model approach lets you build a top-down model of NFV that harmonizes easily with OSS/BSS, and at the same time insure you can fit all the bottom-level stuff into the picture, even if you don’t know for sure what that stuff is yet.
Intent modeling is one thing that unites SDN and NFV because it’s valuable for both and because it’s probably essential in providing a means to adopt SDN to support an NFV evolution. The NFV ISG report makes the recommendation that intent modeling be looked at more broadly within NFV in part because of its obvious value in SDN/NFV integration.
As important as intent modeling is to both SDN and NFV, there may be a still-higher-level issue on the table for both Fujitsu and Huawei, which is virtual networking. We are only now discovering something that should have been clear at least two years ago, which is that simplistic forwarding graphs aren’t ever going to solve NFV networking challenges. The IETF, as I pointed out in a prior blog, has issued a nice paper on that topic, but the real proof of the pudding is the fact that both Amazon and Google have sophisticated virtual networking tools as part of their cloud offerings, and NFV is really functions in a cloud.
Virtualization in any form doesn’t work without network virtualization because you can’t have elastic, agile, resource assignments if every time you add or move something you lose touch with it. Amazon led the industry in recognizing this and adopted “elastic IP addresses” which allowed software to map an “external” address representing access to an application/component to a private IP address that could be changed as needed when something was scaled or redeployed. Google carried this further with Andromeda, which recognized that cloud applications were really living simultaneously in multiple “worlds” at the same time, and that in many cases the same element had to have different addresses and access rules in each.
Neither SDN nor NFV has adopted this notion, and in fact even OpenStack lags commercial giants like Google and Amazon in supporting network virtualization for the cloud. I think that Huawei’s acquisition of David Lenrow and their collateral hire of Chris Donley from CableLabs could well be aimed at taking a leading role in network virtualization. It’s also possible that Fujitsu is heading that way as well, because abstraction at multiple points in SDN is helpful in deploying and managing those multiple virtual-network worlds.
Why is all this happening now? That’s the big question and I think the answer is obvious. Operators have finally recognized that the PoC framework for NFV is proving little or nothing useful because it has holes in the critical places where benefits have to be proven. If you look at VNF on-boarding, just the first step in an NFV service, you see that there are major issues in preparing software as a VNF because even simple points like how you address management elements or how you address dynamically moved/scaled VNFs aren’t standardized. Intent models and virtual networking could fix all of that.
If Huawei could address intent modeling and virtual networking fully, it would have an SDN implementation so good that it would be a real leader there, and that would carry over into connectivity and addressing in NFV. If Fujitsu could abstract any set of equipment under a common intent-model umbrella it could play a role in evolving any operator’s infrastructure toward both SDN and NFV. That’s great for these two vendors, but it’s a bit of an indictment for the others because as I said none of this should have come as a surprise. Maybe what that proves is that the operators are right—vendors are dragging their feet to protect their business model. The only problem with that theory is that some of the vendors, like HP from whom Huawei stole Lenrow, don’t have any network sales to speak of that would be at risk in an SDN/NFV translation. So maybe the issues with other vendors are just a matter of tunnel vision. If so, then these two announcements should widen everyone’s perspective in a hurry.