Light Reading - Telecom News, Analysis, Events, and Research
Sign up for our Free Telecom Weekly Newsletter
Connect with us
Valley Wonk
Craig Matsumoto

SDN & the Commodity Question

November 14, 2012 | Craig Matsumoto | Comments (8)
   
 
no ratings

6:00 PM -- I know a lot of people are waiting for software-defined networking (SDN) to bring down the Cisco Systems Inc. (Nasdaq: CSCO) empire. Waiting for OpenFlow to commoditize network hardware into the ground.

I get asked about that all the time. I just don't see it happening in the short term.

In the long term, I'm less sure.

The seeds are certainly there. OpenFlow allows a controller to dictate commands to switches; ergo, the switches can be dumb and simple, so the argument goes. The possibilities aren't lost on the new SDN companies -- a Big Switch Networks presentation linked to its commercial launch Tuesday repeatedly mentioned "low-cost Ethernet Switches" as its target devices.

The ingredients are out there, too. Intel Corp. (Nasdaq: INTC), at its recent developer forum, released a DIY kit for a commodity server. It uses Fulcrum Microsystems switch chips, Xeon processors and software from Wind River (well, a Linux kernel anyway).

But commoditization of the hardware is not an easy process for the customers, because right now, you can buy a network and have someone else (usually Cisco) do a lot of the work for you. Assembling your own network out of generic Made-in-Taiwan switches might be fun for Google (Nasdaq: GOOG), but it did take a lot of work.

And the switches still have to perform. I like the basketball analogy posed by Simon Leopold, an analyst with Raymond James Financial Inc. (NYSE: RJF), in an SDN report he issued this week: "Imagine how poorly a team would perform if after each pass, the team paused to consult the coach. Rather, the coach provides the general strategy and guidance yet leaves the execution to the team."

There's also the fact that the hardware won't necessarily be cheap. OpenFlow was designed to exploit a router's ternary content-addressable memories (TCAMs), a specialized and costly type of chip. And OpenFlow 1.2 potentially increases the amount of TCAM space you'll need.

So many signs seem to point away from commoditization. And yet -- I can't bring myself to bet against easy and cheap. At the moment, SDN doesn't have either trait on its side, but a lot of people are going to be working to change that.

For what it's worth, the SDN executives whom LRTV spoke to recently don't think hardware vendors are in imminent danger (caveat: two of them are hardware vendors), and Ankur Singla, CEO of Contrail Systems , even cites Cisco and Juniper Networks Inc. (NYSE: JNPR) as being well poised to break away into a software-dominated future. He's got a point.

— Craig Matsumoto, Managing Editor, Light Reading

Newest Comments First       Display in Chronological Order
chuckj
User Ranking
Saturday November 17, 2012 1:29:51 AM
no ratings

You got to love the SDN hardware provider motto:  We are dirt cheap and we sell the same hardware as everyone else using the same merchant chips.  

joanengebretson
User Ranking
Friday November 16, 2012 9:42:21 AM
no ratings

Craig makes a good point about people being used to hand-holding from Cisco. Wonder if Cisco could ultimately follow in the footsteps of IBM as a company that started life as a manufacturer but over time evolves more toward being a consulting/ services provider.

nooser
User Ranking
Thursday November 15, 2012 7:16:41 PM
no ratings

Agree that some large DC operators (e.g Google) have their own controller. However those are not big Cisco customers even today. The challenge will be when Cisco's big accounts are threatened. The broadcom's may sell directly to the DC operators but they are weak on the operations software. The controllers would still have a monetization challenge but it would be interesting to see how that plays out.

big_thinker
User Ranking
Thursday November 15, 2012 6:29:03 PM
no ratings

craig, agreed.... in that case controller vendors would be looking at smaller market not glued to likes of ciscos of the world anyways...

Craig Matsumoto
User Ranking
Thursday November 15, 2012 5:37:00 PM
no ratings

Regarding pica8: Could be. But then again, a lot of bigger customers are accustomed to being handheld by a vendor like Cisco. That could be a hard habit to break.

Craig Matsumoto
User Ranking
Thursday November 15, 2012 5:28:35 PM
no ratings

> under this scenario, how much would a large data center be willing to pay for the controller sw?

I've heard that suggested -- that it might be the controller that gets commoditized.  The context being that the higher-level software (orchestration and the like) was where the big value would be.

big_thinker
User Ranking
Thursday November 15, 2012 2:58:02 AM
no ratings

while vcs are pouring desparate money into openflow and sdn sw startups hoping nicira like returns, they and the startups are working really really (really) hard to make chinese odms like these the real winners...

http://www.pica8.com/

for good reasons they are staying somewhat under the radar and letting the whole ecosystem work for them...  smart.

look at the PICOS, in the link below.  everthing you need in a typical tor is already supported. ovs is there too, so you can do advanced policy pushes in the tors using a controller.

http://pica8.org/blogs/

 

big_thinker
User Ranking
Wednesday November 14, 2012 6:53:26 PM
no ratings

lets say if sdn does commodotize the switch hw, it would be a business model issue for sdn sw companies themselves also.  why?  

broadcom already provides openflow support on its switch chips via their sdk and it will enhance over time and would be available for any controller to use.  when a stable and complete openflow support comes directly from switch chip vendors, odms may be able to sell switches based on these chips directly to mid-to-large data center vendors.  

under this scenario, how much would a large data center be willing to pay for the controller sw?  remember that the largest data center guys may want to write their own controller sw anyways. 

even if an openflow controller company offers a special sw agent on top of these switch chips, how much would they be able to charge for it?  100$, 200$.  how many switches would one need to sell?

we like to point to pc as an example where motherboard and operating system and apps got decoupled.  it worked there because pc volumes were immense and no of kinds of apps that were needed on top of os were bounded only by imagination.  

i argue, it is not the case here and business model is actually a BIG problem for sdn sw companies. 

 

 

 

The blogs and comments are the opinions only of the writers and do not reflect the views of Light Reading. They are no substitute for your own research and should not be relied upon for trading or any other purpose.
More Valley Wonk
Where SDN Is Going Next
6:00 AM Well, it seems to be going everywhere, but here are some guesses as to what might come of that
Juniper Cracking SDN Open
5:00 AM The $176M acquisition of startup Contrail looks like an attempt to create an open-source movement in software-defined networking
Photos: Broadcom's CES Preview
1:35 PM From an 18th-story San Francisco office, Broadcom shared its ideas for a wireless near-future
Buying Into the New Cisco
6:00 AM For all the talk of a software-and-services makeover, Cisco might not be ready to let the switch/router market fall prey to SDN