I haven’t been blogging as much as I’d like – maybe that will always be the case.  It’s hard finding the time with a day job and when each post takes planning, writing, editing, and proof reading.  Going forward, I will try writing shorter posts.  Hope this is the first of many that fits that build.

I find myself talking about SDN more and more referring to myself as SDN optimist insinuating there are many pessimists out there.  Why do I sense that?  I’m a network guy at heart and can’t imagine the past few years without reading about the future of networking and at the same time have learned a lot in other areas of the data center and IT.    It’s time to start doing.  That goes for all of us.  Hopefully those that are “anti” realize controllers don’t have to be all that bad and they can benefit everyone – not just the large scale networks.
So…controllers – they are a great thing for the networking industry.  They can directly impact the way we operate networks.  They can also directly impact the way build networks – but let’s take baby steps.  

How about we focus on network operations and a unified point of application integration first?

Controllers aren’t new. 

In the world of wireless, if one wants to write an application for location based services or RFID, it is integrated with the WLAN controller, not to each AP.  The controller has APIs.  It’s a layer of abstraction.

In the world of voice communication, if one wants to write a slick Unified Communications application for E911 or whatever else, it is integrated with the IP PBX – Call Manager if you will.  It’s a unified point of integration and has northbound APIs.

In the world of server virtualization, if one wants to integrate with applications for DR, storage replication, etc., today it is largely integrated with a hypervisor manager such as the vCenter server in a VMware deployment.  It’s a layer of abstraction and a unified point of integration that by the way, has API integration.

If one wanted to write an application into the network, would we call on human middle ware to try or wish there was an easier way?

Controllers are a good thing.  If fact, they are great thing.  We don’t need to extract all of the intelligence from network devices, but at the very least, let’s extract enough where we have controllers to simplify the management of the network and enable applications to be written on top.  If that goes as planned, maybe we can take the next step and extract *some* control plane protocols.


Regards,
Jason

Follow me on Twitter: @jedelman8
 


Comments

02/22/2013 13:43

My opinion of SDN has evolved .. I used to think it was great. Still think the capabilities enable are pretty amazing. But SDN controllers operating the network is a stillborn architecture. There is too much at stake to centralize the control plane so heavily. Breach, DDoS or forging to control messages from SDN controllers risks too much, in this age of cyberwarfare being waged at the nation-state level.

I think that the ideas are right, new capability, enable programmability, automation of network configuration, but the methods, centralized control, are wrong. I think it is time to change the protocols to embed the capabilities desired and enable distributed control

Reply
02/25/2013 08:21

I can't agree more. My opinion is evolving all the time too. You are spot on, but the point of this post was to start small and take baby steps. We don't need the full control plane on a controller day 1, but rather a controller that 1. centralizes mgmt plane and 2. has a unified view for integrating applications and 3. can incrementally add control plane functions (if desired).

There are many types of networks -- large and small. Each has different requirements, therefore, what is done on the controller can vary (if the technology allows).

Reply



Leave a Reply