Working Thoughts on SDN #2
I was in CA earlier in the week to attend the JP Morgan SDN conference in Palo Alto. To start I would like to thank Rod and the JPM team for the invite. A number of companies presented (Arista, Cisco, HP, Vyatta, Big Switch, Ciena, Embrane, Vello, Cumulus, Cyan and Brocade) and there were a number of startups in the audience: Insieme, Plumgrid and Plexxi. Rather than a reprint of my conference notes, I thought it might be more useful to describe what I thought were interesting discussion points, debates and themes to emerge from my perspective as an observer. I paraphrased the comments from various speakers. My observations below are not exact quotes, but rather thematic summaries of discussions.
Attendance: Most of the startups or young companies sent their CEOs to speak. This includes Arista, Vyatta, Big Switch, Embrane and Cumulus. Juniper was not present. Cisco sent a Sr. Director level person while Brocade and HPQ sent CTO types. I am not sure if there is anything to read into that, but I noted who was participating.
Openflow: There was a lot discussion around the next version of OpenFlow. Dan Pitt (Executive Director of the ONF) and Guru Parulkar (Executive Director, Consulting Professor at Stanford) both spoke on the direction of OpenFlow and the ONF community.
Controller Market Share: Just listening throughout the day I got the sense that Big Switch (BSN), HPQ, CSCO and JNPR (recent press article confirmed by a friend) are all building controllers. I am not sure if everyone’s controller is based on OpenFlow, but Google also built a controller which I am sure it is for internal use and I know they contribute to the OpenFlow working group. I listened to Guido Appenzeller, CEO of BSN, speak and he made some interesting points about controllers. He contrasted BSN with Nicira by saying that Nicira was overlay tunnels for VMs/HyperV, while BSN was focused on three elements of SDN: (i) hardware (HW) switches, (ii) open source Controller which is FloodLight (iii) and apps. He commented that Floodlight was ahead of all controllers and the market (i.e. end-users) is going to want a open source controller and will not be locked into a single controller from a single vendor. The BSN controller is being architected for service providers and enterprises and he believes the SDN market will evolve along the lines of Linux. BSN is not a hardware company, so they are not focused on switches. BSN is focused on the controller for the good of the community and plan to build their business around the applications using SDN. They want to be the App Store of SDN networks. He commented that the first phase of SDN is overlays on the existing brown field networks, but soon there will be a new hardware/switch layer that enables the second phase of the SDN evolution. In a later panel, Jayshree of Arista commented that there might be application specific overlay controllers.
App Stores for Controllers: When I hear people describe this view of SDN it always involves applications like load balancers, firewalls and routing. I think of these things as network services and I do not think that network services belong in App Stores and I am not convinced they have anything to do with SDN.
When Does SDN Matter: Everyone agreed that SDN is starting now, it will accelerate and reach critical mass in three years which is the start of the sustained period of adoption and growth years. Dave Stevens, Brocade’s CTO commented that SDN does not reduce the power of incumbency and that the big stacked networks that have been built over the past twenty years are complex and really hard to switch out. I would comment that this is exactly the problem statement for SDN. He commented that SDN is about automation and that many of their customers are asking what does SDN mean to them. Brocade’s plan is to partner for SDN.
VXLAN: There was lots of chatter on VXLAN. I understand it, I just do not get the excitement. Wrapping something in more of something is not really innovation. If you get excited about VXLAN, you probably think fondly of the following terms: Multi-bus, IGRP, AGS, RIP2 and you have not missed a standards body meeting in a few decades.
SDN in WAN: There were three transport focused companies at the conference: CIEN, Cyan and Vello. Ever since the Google OpenFlow WAN presentation at2012 ONS, there has been a lot of interest in how SDN fits in the WAN. The VP of Engineering for Vello commented that transport is complex, resources (i.e. fiber?) are scarce and cited the Google OF WAN. They are focused on DC to DC interconnects. The CTO of Ciena (an old colleague from my CIEN days) commented that the problem SPs have is how to get a lot of capacity to many locations cheaply. He describes this decade as the M2M decade coming off the mobility and internet booms from the prior decades. He also made some other comments that I would agree with: (i) if SPs can separate compute of topology from path calculation that could become very powerful for carriers and (ii) the SDN proposition for the carrier is the ability to have levers that change the underlying wires of the network more than once every 3-7 years. Imagine if they could change them many times a day! That last sentence was a veiled Plexxi comment by me.
SDN for High Capacity Infrastructure: One of the speakers I was most impressed with was J.R. Rivers of Cumulus Networks. I had never heard him speak before. He was part of a SDN panel with three other companies, but had time to make some interesting comments. He said that Cumulus was focused on high capacity infrastructure. He talked about building DCs in which servers were made up of communities of systems and the relationship of how these communities peer with other communities in the DC. He cited a example of how a community of ten servers might work together and that there are many apps in a DC and that clusters of machines want to work with other apps in different clusters. He thinks SDN is a way to deploy apps in the DC and that enterprises are in the proof of concept phase for SDN. He commented the SDN might be the SLA for the application in the enterprise. He does not think that SDN obsoletes hardware, but rather SDN allows the network to run different parts of software on different end points. SDN is not path computation, it is connectivity and SDN enables virtual and physical machines to communicate and in totality this becomes an enabler for SLAs to move into the overall network system.
Panel on Switching and Networking: There was interesting fireside chat with Arista, BSN, Brocade and HPQ towards the end of the ay. Jayshree of Arista views SDN as a seminal moment in networking where the orchestration of the network is improved and this translates into OPEX savings. Guido commented that controller architectures will evolve and we will see an evolution of the first generation centralized controllers to a distributed controller model. There might be one logical controller and a distributed set of computational controllers like a hadoop cluster. There was a discussion about the development of controllers of controllers and application specific controllers for overlay functions. At the end of the discussion of controllers, there was a discussion that there was no good solution for the controller of controllers and that this was probably a good PhD thesis to be written. There was unanimous agreement that the VSwitch is a strategic point in the network. With all the SDN talk, HPQ pointed out that not everything is about SDN and managing the legacy network is just as important and customers want a single pane of glass to manage the network.
Cisco Systems Meme: Throughout the conference there were discussions on the future for Cisco and what will be the affect of SDN on Cisco; this is kind of what happens when you run around the world with a presentation entitled “Cisco and the Seven Dwarfs” from eight years go. This has not been a topic I have been shy about, but there was a neatly packaged thesis that made sense to me from several of the CEOs of SDN startups who were former Cisco executives. The thesis is that Cisco has grown up to be IBM. They develop their own ASICs, they package their ASICs/Hardware with their closed software and sell it as a system at very high margin. Think IBM mainframe. The only way to beat Cisco is to be cheaper, better and faster. This was clearly the strategy of Arista. Along comes SDN and it is going to unleash the value that is contained in the closed system. I once wrote a book on what happens when you deregulate a market called telecom and clearly many of the SDN startups are looking to deregulate Cisco’s market share. The companies that are building SDN solutions see Cisco as the IBM of the 1980s and they want to unlock the value in the closed system. In one way this makes sense to me as I started my technology career selling multi-protocol bridges and routers into SNA networks. I was told as a young sales person that no one gets fired for buying IBM and to beat IBM you need to be cheaper, better and faster. I know today that no one gets fired for buying Cisco. This was reinforced just yesterday when I got an email from a CEO of Cloud Provider who told me he was not interested in any alternatives to Cisco because they are a Cisco shop and Cisco now has SDN. The caveat with this story making sense to me is I might be falling victim to the power of stories and legends. 😦
Great summary of the JP Morgan conference. Thanks!
Can you tell me what you mean by “SLAs” in Cumulus’s comment: “SDN is not path computation, it is connectivity and SDN enables virtual and physical machines to communicate and in totality this becomes an enabler for SLAs to move into the overall network system.”
Service Level Agreements? Trying to understand the finer points in this interesting comment by Rivers.
JR is probably the best person to ask, I was just summarizing what he said, but I think what he meant was tightly coupling the physical underlay with the overlay network.