Here is something I have been kicking around in my head for some time, and wishing for a quick answer…
In the networking world everyone has been up in arms about OpenFlow for a while now. Some thought leaders in networking think that OpenFlow (Software Defined Networking) has a real potential to shift the architecture of the network (in other words it’s not vaporware). I am no (OpenFlow) expert, but to me it seems like the goal here is to flatten the network, making it easier to manage, scale, and deploy. But OpenFlow is really a protocol and nothing more. There is no structural advancements. By structural I mean our structured cabling and power infrastructure. Business as usual in those parts of the building. IDFs are still packed with access layer switches (ASICs). The OpenFlow crowd would not agree with me here (probably rightfully so), but to me it is applying the principal of controller based wireless access points to multi-layered networks (read: layer 2, layer 3, access, distribution, core).
Now, Motorola has been quietly marketing this little thing called Passive Optical LANs. These bad boys are a progression of the network carried over from service providers who have used Passive Optical Networks (PONs) for some time now (think: fiber to the home). Sometimes these are referred to as GPONs (Gigabit Passive Optical Networks). Motorola is marketing this solution for enterprise / campus networking rather than just carriers. Before going forward, stop and consider: Google purchased this division of Motorola. Let that sink in…
Now it seems to me if the goal is to centralize the network, drawing the intelligence back toward the “physical core” (I use my words wisely there), then POLs would be a superior choice over something like OpenFlow. After all, why not scrap the need for ASICs all together rather than just separate control from forwarding.
There are drawbacks to POLs. For one, it’s basically “fiber to the desktop” reinvented. While this reduces the need for switching IDFs, it does add an additional need for unique access layer hardware / media conversion.
So here is what I am driving at – looking at business, operations, and technology drivers I would suggest POLs should be considered more of a “threat” to the traditional networking models than OpenFlow. Yet, it’s OpenFlow that gets a symposium. Now, I am no conspiracy theorist but think about this: OpenFlow permits the continued sale of hardware (data/forwarding plane still consuming building power in IDFs), while also doing something to move networking toward a more efficient architecture (not power efficiency, but operational/management efficiency). Meanwhile, if POLs took off, all those 3560s, 2960s, and 3750s stand to get trashed…literally. And because our industry is driven mostly by a small group of vendors, why would they eat their children in the name of advancement? Ask yourself, why are vendors backing OpenFlow?
I may post more later, as I am slated to meet some clients who chose to use POLs in their corporate infrastructure. I am not sold, but my mind is open (always).
Some interesting discussion takes place here: https://plus.google.com/u/0/116575301739886800473/posts/QQLbvkknNbe
Really interesting thought regarding Passive Optical LANs. Now there’s a truly disruptive technology!
As the organizer of the OpenFlow Symposium, by the way, I can assure you there’s no secret conspiracy. Just genuine interest in challenging the state of the art in networking!
I’ll let the folks who know networking more than I do weigh in next…
[...] OpenFlow and Passive Optical LANs [...]
Now this is interesting…. Let me kick it around the cranium for a day or two…
POL (particularly GPON) is multipoint SDLC reinvented. It might be a good idea for carrier access network (particularly because you don’t need active devices “in the wild”), but you don’t want to see that in your data center, trust me (or ask anyone who had to deal with SNA or SDLC 25 years ago).
Not data center, enterprise LANs or traditional building and campus LANs…?
PON is a “last mile” access technology, i.e. it is inherently point to point. Here is a good technical overview for those unfamiliar with it: http://www.scribd.com/doc/35992488/GPON-Fundamentals-20070606-A
The typical set of requirements a PON deployment is addressing are: (a) use as little fibre as possible to reach as many end-user premises as possible; (b) deliver reasonable amounts of bandwidth as cheaply as possible; (c) enable triple play services (Data, Voice, Video) to these premises. Video can be either over IP or more traditional RF, over glass (RFoG).
There is not much overlap between what PON has been designed to do (including the case in question by Motorola), and what OpenFlow is targeting.
Multipoint technologies, such as PON, have all sorts of limitations. For applications with limited use cases such as last mile, it makes sense. For a data centre or an enterprise campus, multipoint protocols and cabling cause all sorts of indeterminate states in the propagation layer – and this uncertainty creates poor adoption.
In the end, the network is a foundational technology that underpins an IT Infrastructure. Any level of uncertainty means it would not be accepted.
OpenFlow/SDN and PON have virtually no intersection in my view. Different solutions to different problems.
POL != PON
I will get more raw data and architectural design specs and post later. I met with a customer yesterday who said he would not have flinched to convert had he not invested 18 months earlier in a new edge layer.
It is interesting the backlash though…
I stand by my comparison of OpenFlow to POL. This is not an “EIGRP vs OSPF” argument. This is more of a higher level architecture debate about how we do enterprise LANs, period. OpenFlow is the traditional model, but trying to advance this existing architecture to fit more modern needs. POL does the same thing, but with a new architecture (new to LANs).
Stay tuned for more posts…..this topic has traction.