• 11/19/2013
    9:06 AM
  • Rating: 
    0 votes
    Vote up!
    Vote down!

SDN Upends The Switch & Router Market

Software-defined networking (SDN) is changing the networking market as we know it. In the midst of the confusion, it's clear that open source and standards-based networking are gaining ground.

Every now and then, the industry seems to say, "OK, enough of this; it's time for a change." It's a sort of an antitrust action against the status quo because the needs of the market are not being met in economical and creative ways. The hype around SDN is so loud because it promises to upend the switch and router market, of which Cisco is the kingpin.

It's not surprising, then, that Cisco, which had previously signaled its distaste for SDN, has responded with the much-anticipated unveiling of its Application Centric Infrastructure (ACI), a datacenter network solution from its spin-in Insieme Networks. The key pieces of the platform are a new line of Nexus switches and an Application Policy Infrastructure Controller (APIC) -- which one observer aptly labeled as Cisco 's "un-SDN" answer to SDN.

But is it just bad timing that the very day the industry was abuzz with the ACI announcement, Dan Pitt, head of the Open Networking Foundation (ONF), explained publicly why the organization did a sudden about-face and agreed to develop northbound interfaces (NBIs) so that the brains of an SDN infrastructure, the controller, can talk with other parts of an SDN solution, such as applications and management software? To be sure, the ONF, which has done much for SDN by developing the standard southbound interface, OpenFlow, had good reasons for holding off on the northbound side, chief among them to see if the vendor community itself could agree on NBIs.

The lack of clarity on what shape SDN eventually will take is bound to make prospective customers hesitate. Take something as core to SDN as the controller. It does not help that Cisco and Juniper, both platinum members of the OpenDaylight project, the forum working on an open-source controller, have distanced themselves from the group's work. Cisco is taking the position that while its new Nexus switches will work with OpenDaylight controllers and OpenFlow, it would be better for customers to go whole hog on the fully-Cisco ACI platform, rather than trying to integrate the open-source technology. Juniper even hinted that it may not have any use for the work of the OpenDaylight consortium at all.

At the same time, the ability to implement so-called "bare-metal" or "white-box" switches will challenge vendors like Cisco and others who sell closed boxes. This new breed of switches could do to hardware what SDN does to network infrastructure. Bare-metal switches disaggregate switching in the sense you have only a box based on merchant silicon, as opposed to one using custom ASICs. In other words, the switches have no operating system or any software that would utilize the operating system. Some of the biggest names in the business -- Google, Facebook, Amazon, Microsoft -- are reportedly buying generic switches directly from original design manufacturers and bypassing vendors like Cisco, Juniper, HP, Dell, and Force10.

If the idea behind SDN was to break open the proprietary switch and router market and encourage innovation, the argument could be made that it has succeeded spectacularly. In this regard, Cisco is to be thanked for continuing the debate on the new network paradigm and offering customers more options than they perhaps might want in having to choose between an un-SDN Cisco ACI solution and various-source solutions. And on the open controller front, we have OpenDaylight developing a northbound interface, while Juniper offers its own open-source controller.

In short, the choices for the new network are clear as mud. Going forward, customers can opt for Cisco, open-sources (plural), and/or standards-based networks -- at least until the market sorts things out.



SDN chaos

Sam, thanks for a great analysis of the chaotic SDN market. Do you think all the tumult wil cause some customers to go with Cisco's "whole hog" solution with a desire to move forward and the philosophy of "better the devil you know"? Or will most simply wait out the development stage? It seems it could be quite a while until things stabilize.

Re: SDN chaos

This echoes a comment I made here in the last day or so, so I'm in agreement with you here! At the moment, if it's possible, we have too many options. If I might pick out one sentence for ambiguity's sake:

"Going forward, customers can opt for Cisco, open-sources (plural), and/or standards-based networks -- at least until the market sorts things out."

The implication here seems to be that Cisco isn't open-source or standards-based, but the waters are muddied by vendors supporting supposedly "open" APIs. A white box switch isn't necessary "open", but if it supports a desirable API (e.g. OpenFlow) then does that make it an open solution? If Cisco's APIC can use OVSDB and Netconf (among others) southbound, is that "Open" enough if we assume that the intelligence behind it can be proprietary? 

Re: SDN chaos


Can't really argue with you because I agree with you (well, that was easy). I think we have some ways to go before the market decides what it wants by way of "open." Frankly, at times I wonder why industry groups don't just define a lot of APIs (north/south) and leave it at that. Like a lot of people, I'm waiting to see how things play out...

Re: SDN chaos

As a side note, my own vision for how SDN will go is that in networks that choose to adopt it, we may well end up with a hybrid solution. e.g. we may choose NV so that we can have inifinitely (allegedly) scalable firewall service provision (especially in multi-tenant environments) perhaps even with optimal placement of the service, but then the network itself may be managed by another controller which will optimize the flows themselves. We're not likely to get away from VxLAN host encapsulation either, and some smarts between, say, vCenter and a network controller would be good (or do we go for the full NSX solution?).

Perhaps the underlying compute and network hardware will be white boxes, and perhaps not, but I fear for compatibility and support. The first time a white box doesn't behave as expected when programmed by a controller from another vendor the finger-pointing will begin, and each vendor will say the other has mis-implemented the (rapidly developing) specification. If so, these APIs wouldn't be the first set of "standards" that fail to interoperate. The security of buying an integrated solution from a single vendor will be, I believe, a strong selling point, from a support perspective if nothing else. And let's not even speculate about which vendors will, uh, 'enhance' the standard to provide additional features...

Re: SDN chaos

jgherbert, your vision of the hybrid implementation makes a lot of sense to me. I can see having multiple controllers for different capacities, especially the underlying network itself.

The dynamic between open and proprietary is almost comical in IT. It seems like SDN itself is a rebellion against proprietary hardware. But then no one can agree on what is really open, so allt he vendors create their own version, and we don't really think generic hardware is good enough to use anyway.

I guess that's what allows Cisco to release an "open" solution but advise customers against implementing it without creating too big of a stir. Oy veh!

Re: SDN chaos

Your second paragraph is why I don't think folks can just write off Cisco. Another thing I'm not hearing anything about is testing and certification of SDN solutions.

Re: SDN chaos

Final thought - sorry about this - regarding the title:

"SDN Upends The Switch & Router Market"

We should probably add to that in our minds "& Firewall & Load Balancer & Network Monitoring Market". If you're in the business of selling FW and LB appliances, you'd better be able to virtualize (and do it effectively). 

Network Monitoring is something that isn't mentioned much, but the potential for some amazing real time monitoring exists in the SDN world in ways that just wouldn't be possible currently. Let's face it, SNMP is not the ideal mechanism for real time monitoring, yet that's the standard we have been using for, well, too long. SDN opens up a whole bunch of new possibilities.

Re: SDN chaos

jgherbert, you caught us on that one! I think we also would need to add data center and cloud, and on and on. But then you would never read the whole article, would you? :)

Re: SDN chaos

Hi Susan,

The end of your post pretty much says it--yes, I think it will be quite some time before the market gives a clear signal whether it likes the pure Cisco strategy or wants the open SDN model. SDN claims to offer a very clear capex advantage, but that alone probably will not be enough to really challenge Cisco. Besides, SDN also has to prove itself from a reliability, performance and scaling perspectives.

Anyone who thinks that Cisco can be shunted off to the side is dreaming because Cisco has the bucks and the customers, and I don't see it rolling over and playing dead. Can anyone see Mr. Chambers go out this way?