Networking Thoughts to Start 2014
SIWDT is my second blog. My first blog lasted few years from 2005 to 2007. I created my first blog, which can still be found on the Wayback machine, after having an unexpected breakfast with Dave Winer in 2006. It was at a conference in San Diego and if you do not know who Dave Winer, please stop reading now and go back to whatever you were doing before you thought to read this blog.
Life has a way of interfering with writing. I hate that about life and that is why I have immense respect for people who can consistently write. The late Tom Clancy comes to mind, JK Rowling and Dave Winer. Here we are in Year 3 or 4 of SDN. Who cares really? I do not care about SDN, what I care about is building a company and to build a networking company today, the company must do something different, unique, that is technologically apart from the rest of the players. That does not mean the company will be successful, it is just the first step. I spend my days talking to people about networking and couple of weeks ago I had two orthogonal meetings.
The first meeting was with mid size enterprise company. It was as if I had stepped back into time to talk about VoIP, SIP and client/server. They were building a new DC and they had no clue about using a controller architecture. Their network was 100% Cisco, every piece of EMC equipment and splattering of NetApp. One of the lead network architects told me that he had told the CIO that “Plexxi seemed very futuristic” to him. I have learned that when I encounter customers trapped in mid-oughts or the 1990s, the prudent course of action is to thank them and leave quickly.
The very next day I visited a service provider. I am purposely disguising the customer description. I was attending the meeting alone as I thought I was going to spend an hour in the office of a network director and I figured I could handle an hour meeting on my own when the real objective was get the next meeting. I soon found myself in a room with five network architectures. What a difference a day makes. We talked about controller architectures, failure domains, chordal rings, WDM, traffic loads and building high capacity multi-site fabrics. These were my kind of people.
I have been working in the SDN part of networking for a few years now and I can state without question that building a high-performance controller for physical and virtual networking is hard. If we are going to believe all the scale up by increasing compute densities, deploying flash in the core, moving to 100G silicon photonics and all the other stuff we dream about at conferences, some company is going to have to build a high performance controller that supports a >1 billion calculations a second. Please don’t talk to me about controllers doing 250k a sec. In a few years, I am going to be sitting at a conference and a speaker is going to put a chart up showing five years of controller development history and this chart will show every increasing controller benchmark speeds by a few companies. The chart will look like a history of CPUs or the development of jet fighters.
FYI: that is the real game that is being played. The most impressive customers I meet are the ones that start the meeting asking about our controller. Why care about ethernet switches? We are all using the same silicon and operating on Linux. We are all building various DevOps hooks. Ask me what a Plexxi controller does and I will tell you that it computes efficient photonic topologies based on workload and workflow needs.
Plexxi is >2 years into the development of our controller architecture and it is not without triumph and tragedy. I estimate that the Insieme team is ~18 months into the development of the APIC controller. I have no idea where in the development cycle Arista, HPQ, Brocade, DELL and Juniper are, but I have seen the APIC controller pitched competitively in the market.
I wrote this post seventeen months. I forgot about until I was reminded by @RealLisaC eight days ago. Plexxi is three years old this month. We have a long way to go and building something different is not for the faint of heart.