Networking - Hot or Not with SDN ?

Software Defined Networking or SDN has become an overused term in the industry today. Companies, big and small, want to be associated with this new technology wave, because they do not want to be left out, when the wave hits new highs. Every day, you might come across companies that claim to offer SDN products or talk about their products being "SDN compatible". Often times, these products have been selling and shipping way before the term SDN was even coined. One thing, however, is undeniable - Networking has become hot again. And within all this hoopla, there is truly some disruption going on in this industry -- disruption in the way networks are being created and managed; disruption in the way functionality is being abstracted out of the physical networking devices; disruption in the way astronomical amount of network state information is being stored and analyzed; and disruption in the way new business models are being created.

In this post, I try to highlight who are the main players in the SDN arena, and what are some of the key M&A activities that are fueling the growth of this sector. I also try to highlight some of the smaller players, who have innovative products and are likely acquisition targets.

But first things first -  What is SDN?

The most commonly used and simplistic definition of SDN is that, it is a way of architecting a network by separating the control plane from the forwarding plane, centralizing the control plane, and keeping the forwarding plane distributed, and providing a single pane-of-glass management, and in addition making the network programmable.  (If you are new to these planes, just remember these are conceptual planes rather than physical ones. Click these links to understand more about the control and forwarding planes). That actually tells more about how it is implemented rather than answering the simple question of what it is. In my mind SDN is offering Agility and Automation in the network -- Agility in terms of setting up and managing the network and provisioning network and security services; and Automation in terms of how you can do all of those programmatically.

Now let's go back to the implementation of it. In order to do that I like to look at it in a layered-cake model - "layered" in terms of layers of abstraction. 

• At the very bottom, you have the physical network consisting of physical & virtual switches & routers, physical & virtual load balancers, physical & virtual firewalls, etc. This is for the most part quite immovable, inflexible, distributed, and a lot of this layer is still very manually provisioned. This is referred to as the 'Underlay'. 
• On top of the Underlay you have ... (no prizes for guessing, Yes that's right) ... 'the Overlay' network. This is the network virtualization layer that makes the overall network more dynamic and flexible. It also removes some of the constraints and limitations that the Underlay Network has (e.g. the 4K limits on the number of VLANs you can have in a given network). You can think of the network virtualization layer as a sort of virtual data cable, which can span different physical networks across geographies, can be moved around quite flexibly, and can be installed programmatically to connect different devices. Tunneling technologies like VXLAN, NVGRE, STT, etc and some proprietary implementations such as Juniper's JunosV App Engine, Cisco's vPath, etc.. all support this layer. This is one of the key elements of the SDN, and a number of M&A has happened, and will continue to happen in this layer.

Figure: SDN Layers of Abstraction

The Overlay and the Underlay are together referred to as the forwarding plane (a.k.a. data plane) of the network. Given that several network services such as firewalls, load balancers, etc. run in this layer and there is physical transport of data here, terms like 'Services Plane', 'Transport Plane' are used by some companies to be belonging here.

• On top of that you have the Controller layer. This is a logically centralized component whose job is to understand the state of the network, analyze the state, and then change the network state as per the instructions it gets from the layer above. Realize this, that there is an astronomical amount of network state data that gets processed in this layer, and so this has to be architected in an extremely scalable and resilient fashion. Southbound this layer communicates via different standard protocols, Openflow, of course, being one of the most talked about ones. Again a lot of research, innovation and M&A has been happening, and will continue to happen, in this layer. 
• On top of the Controller you have the Orchestration layer. This is where the controller gets its instructions through configuration and based on those, and its understanding of the state of the network, goes ahead and issues policy commands to the network. 

The Controller and the Orchestrator together make-up what is the Management & Control Plane.

• And finally there is the Application layer. The application layer could be interacting directly with the controller or with the orchestrator, but here is where innovation can hit the sky. If you have APIs and the whole network is in your hand, imagine the possibilities..

If you pause for a moment and think about it, separation of control and data traffic is not a new concept - networking companies like Cisco and Juniper have been doing this all along. It is just that SDN does it in a different and unique way and more importantly in an open and standards-based fashion, where different companies (big and small) can now partake in this ecosystem. 

So what started all this ?

Well, there wasn't a single trigger point for SDN. Sometime towards the latter half of the past decade, a few things started to occur, almost in parallel.

1) Traditional networking companies like Cisco, Juniper, Alcatel-Lucent -- started opening up their networking platforms by offering APIs (Application Programming Interfaces) to program their forwarding plane. These companies had been developing L2/L3 devices and making their margins on expensive hardware. They soon realized that with the world moving towards software, being a hardware-only company would only work to their detriment; case in point was Sun Microsystems, which, without a solid software strategy, went from being one of the hottest to becoming almost irrelevant. So these networking companies began offering software APIs and development kits (Junos SDK by Juniper, Application eXtension Platform by Cisco, AAPP by Alcatel, HP Alliance ONE by HP, Arista's EOS APIs and so on) and began encouraging higher-layer network services to be developed and deployed on these devices.

2) A group of researchers at Stanford and UC Berkeley began tinkering with the idea of using an open protocol namely Openflow, to control network devices. It was this group of researchers that later coined and popularized the word SDN. In fact, for a long time, SDN became synonymous with Openflow. Later on, an organization was formed, namely Open Networking Foundation (ONF), that promoted the research and also invited thought leaders in the space to talk about SDN and Openflow and promote them. A number of companies like Big Switch Networks, became some of the early adopters, and began developing products and solutions supporting Openflow. The biggest success out of this was a company called Nicira, founded by one of those original Stanford researchers namely Martin Casado, who wrote the first draft of Openflow while a PhD student at Stanford. The company as we all know was later acquired by VMware for a sweet $1.2 billion. And that .....  caught everyone's attention...

3) In 2005 Intel began shipping its first dual-core x86 processors. From there it went to quad-core processors, and then to eight-core Sandy-bridge, followed by its 10-core Ivy-bridge (still to hit the market). This trend definitely brought in a lot of attention of the networking community to the capability and potential of x86 processors to do networking related processing. Folks in the networking world suddenly realized the excellent price-performance ratio of merchant silicon vis-a-vis expensive custom based silicon that traditional networking companies built and used., And they suddenly began exploring the possibility of using cheap hardware to run networking software. Later on, Intel also started offering its Data Plane Development Kit (DPDK) and companies like 6Wind, and WindRiver started offering entire stacks with Intel DPDK for faster networking traffic processing. 

4) And finally one of the things which SDN definitely owes its occurance to, was the open-source community gaining a lot more steam especially with The Linux Foundation getting formed in 2007 with the merger of OSDL (Open Source Development Labs) and FSG (Free Standards Group), and in the same year KVM Hypervisor coming to the limelight with Red Hat's acquisition of Qumranet, the maker of the KVM hypervisor. The impact was so much that several "closed" companies began actively participating in the open-source community.

Through these few apparently-disconnected set of events, people suddenly had access to cheap hardware, free OS and virtualization software, and a good open-source backed networking community, and for good measure the networking incumbents also agreed to open up their platforms for innovation...And that, my friends, set the stage for the SDN revolution.  

Lay of the Land

The SDN and networking technologies are becoming more dynamic and fast-moving than ever before, and so is the industry that is driving it. Companies are being spawned (just like virtual machines) in no time, because they have access to so much VC capital; and then are being acquired in no time because today, buying as opposed to building is a differentiating strategy that all companies (irrespective of whether they have the M&A genes or not) are adopting, to keep pace with the technology innovation. And this is spurring more innovation, more start-ups, more VC funding and in turn more buying...and the cycle of growth continues... 

So let's look at the players in this space and, while we are at it, also look at the M&A that has been happening here. Again its best to look at them from the layered-cake perspective - and just to switch things around, let's start from the top.

Application Layer

I feel that the bulk of innovation is going to happen here, but mainly after SDN matures a bit. Applications will be re-designed to take this new networking paradigm into consideration. The Controller northbound APIs will enable these applications to do things they have never been able to do before and that will definitely spur innovation.

• Among the big players, Cisco has announced a few applications such as Network Slicing, Network Tapping, Custom Forwarding; Juniper talked about applications such as bandwidth calendaring etc., which reserves network bandwidth and offers it on-demand; HP announced HP Virtual Cloud Network application which enables cloud providers to create scalable and flexible Enterprise Clouds. 
• vArmour offers security service in an SDN fashion. They call is SDSec (Software Defined Security) and the idea is that instead of creating a standalone firewall or security service, they have the control plane and the data plane of the security application segregated.
• Big Switch Networks has announced a couple of applications namely Big Tap (Network Monitoring Application), and Big Virtual Switch.
• CPlane offers a couple of applications such as 'Discovery and Correlation' (automatically discovers VMs, Servers, etc.), Traffic Engineering, Layer 2 VPN, and Layer 3 VPN.

Many of the applications (or application use-cases) are still in an early proof-of-concept phase, and are really not shipping for revenues (although Big Tap and Big Virtual Switch has $ price attached to them). There will be ISVs probably developing SDN applications. However, as of now there aren't many companies solely focusing on SDN apps, probably because there isn't still an agreement on standardization of the Controller Northbound APIs, and the SDN market is still relatively immature.

Network Orchestration Layer

Most of the current players in this layer are companies that have existing server/cloud orchestration systems or  network management systems and are re-purposing those for SDN based network orchestration. In play here, is a good mix of open versus proprietary technologies, as well as in-house developed versus acquired technologies.

• OpenStack is, of course, the most popular one, mainly because its open-source and it has backing from the community. The Openstack community is quite active and has helped enhance this to a point where there are now it can be used for cloud orchestration. It is designed in a modular way and has different components handling different areas of cloud orchestration. There's a networking component called Quantum, and a compute component called Nova, which are probably things that very relevant and most folks start out with these. Then there are other components for storage, image management, dashboard, identity, etc. There is also ability for plugins from different companies for other functions such as load balancing, firewall, etc. Nebula is a well-funded start-up that is providing a packaged version of OpenStack on a 4U x-86 box - they call it Cloud Controller (not to be confused with SDN Controller). Nebula's CEO, Chris Kemp, was one of the progenitors of OpenStack.
• CloudStack is the orchestration system offered by Citrix. This was their $200+ mm acquisition of Cloud.com that Citrix open-sourced. There is debate as to which one (OpenStack or Cloudstack) is better. Interestingly, Citrix participates quite actively in the Openstack community as well, mainly to have their Load-Balancing plugin to be adopted widely.
• VMware vCenter, IBM Smartcloud Provisioning, HP's Intelligent Management Center (IMC), and Micrsoft System Center are existing provisioning, management and orchestration tools will be used for the SDN-based network orchestration as well.
• Then there are other smaller companies such as Chef, Puppet Lab, Eucalyptus Systems, etc. that also offer orchestration systems.

This layer also covers some of the aspects of network management, analytics, visibility, and troubleshooting etc. A number of companies are re-purposing some of the analytics and performance management products for SDN.

• Riverbed recently announced additional capabilities in its Cascade product specifically to have visibility into VXLAN traffic. A related acquisition that Riverbed made late last year was of OPNET for $1 billion. Cascade product is supposed to have some integration with the OPNET's application performance monitoring (APM) products.
• Boundary is another analytics company that recently made a move to offer SDN solutions. It also has a partnership with Plexxi on that front.
• Guavus, an analytics started off as a generic analytics company but are now gearing some of their products towards SDN.

SDN Controller Layer

As I mentioned earlier, this is the layer where quite a bit of innovation, M&A, and partnerships have happened and are on-going. I feel there is going to be a good number of acquisitions here especially for big players wanting to diversify into the SDN space or for companies to expand their SDN controller capabilities.

Here are some of the SDN Controllers:

• Cisco announced their SDN Controller. Their ONE Controller is a home-grown one (quite unlike Cisco !), but I think that they are going to acquire a company or assets to beef up their controller functionality. Probably Insieme ? 
• Juniper's controller story was enhanced through an acquisition of Contrail in Dec 2012 for $176 million. The interesting thing was that it came out of stealth mode only a couple of days before its acquisition.
• IBM has announced its own DOVE (Distributed Overlay Virtual Ethernet) controller late last year. It also put a price tag of $100,000 for a license, clearly indicating that this is not for the small and medium sized enterprises or data centers.
• NEC's ProgrammableFlow Controller is also something they have developed in-house and has also won several accolades at Interop 2012.
• HP announced its Virtual Application Network (VAN) SDN Controller to enable SDN solutions for Enterprise.
• VMware uses the Nicira controller for its network control functions.
• Big Switch Networks has been doing some of the pioneering work on SDN controllers especially using Openflow and is a likely acquisition target for anyone trying to get its own controller. 
• Other notable players are Cyan and Adara Networks. Cyan, a optical networking gear company made a move into software and created Blue Planet, a controller that works with different networking vendor switches. Adara has done quite a bit of SDN work, and has created its Full Stack Engine (FSE) SDN solution. 

As you can imagine there is an astronomical amount of state information that the controller has to maintain and analyze. So analytics and big data becomes an integral part of the Controller. Some products have the analytics component embedded within their Controller, and there are others using analytics engines from 3rd-party providers.

• IBM, of course has a significant investment in Big Data, and so their partcipation in the controller does not come as a surprise.
• Hadoop and Apache Cassandra are products and components that are being more and more relevant in the SDN world for many to develop their own analytics engine for the controller.

And finally there are other network optimization SDN products which fall in this layer as well mainly because they are controlling the underlying network in some way or form and providing use-cases such as bandwidth calendaring to provide network SLAs and demand path placement. Cariden acquired by Cisco in Nov 2012 is such an example.

Network Overlay or Network Virtualization Layer

As I mentioned before, the main job this layer does is creates the so-called "virtual data cable" that can span geographies and connect devices (both physical and virtual) across different networks. The reason I use 'data cable' as an analogy is because the upper layers want to have seamless L2 access to the network, i.e. for them the network should look like an L2 network. So how do you get an L2 access that spans geographies, or in other words how do you get a cable to connect two different devices that are separated by L3 networks? Well, one of the answers is, of course, through tunneling. By encapsulating packets with L3 or L4 headers, you essentially use the existing (underlay) network to transmit packets to different geographies, and at the other end, you decapsulate the packet and read its contents and ..Voila! you have a L2 data cable that has connected two devices in 2 different networks in 2 different geographies. In essence  you used the existing infrastructure to broaden the perimeter of your network by making it a flat L2 network.

Note that all the devices participating in this overlay network, should be able to effectively extract the packets out of the tunnel, and so there are several.common open standards-based tunneling technologies that are used in the overlay such as VXLAN (developed by Arista, VMware, Cisco, and others), NVGRE, STT, etc. There are numerous other ways you can tunnel packets to achieve the above goal such as IP-on-IP, UDP-on-IP, etc. SPB (or Shortest Path Bridging) is something that Alcatel-Lucent uses.

Players in this space again include small as well as big players:

• Nicira (acquired by VMware) is, of course, one of the biggest players here. They were the pioneers of network virtualization and also popularized it
• Cisco offers ONE / vPath overlay on their Nexus 1000V products. Cisco's products support VXLAN based overlays because their own technology OTV has a packet format strikingly similar to VXLAN. (Not surprising, because Cisco had a big contribution in formulating the VXLAN technology). Note also that Cisco also acquired a company called vCider which created private cloud segments within public cloud using overlays. They productized this acquisition through their Intercloud offering.
• Juniper offers their JunosV App Engine creating tunnel overlays on top of physicial networking products.
• HP offers Ethernet Virtual Interconnect (EVI) which is an overlay technology that provides connectivity between data centers.
• IBM offers its DOVE network that allows for flat cross-data center network.
• Midokura's MidoNet, creates an overlay on top of generic IP fabric. Midokura is a Japanese company that entered the US market last year and is another very hot acquisition target. Midonet has integration with OpenStack, and the company is very active in the OpenStack community. Here's an article that talks about the product in quite detail. Interesting to note that in their approach, there is no central controller. Its all distributed within the MidoNet agents.
• Contextream's Grid offers a software abstraction layer on top of traditional networks. 

Cisco's ONE / vPath and Juniper's JunosV App Engine also have this concept of a "service chain", that creates a logical order in which packets traverse these network services and get serviced - e.g. Firewall chained to Application Load Balancer - You want your traffic to first hit the Firewall where you drop malicious traffic and then you send the traffic down the chain to the Load Balancer to service it. This could arguably within the Application layer, because it is a use-case that is being enabled using SDN. However, I'm listing it here, because "Service Chain" needs some implementations in the overlay layer.

This layer should see more and more innovation and M&A, specifically because of the different use-cases it enables that overcome some of the limitations that the physical networks have.

Network Underlay Layer

This is where a wide variety of players exist. Starting from big routing & switching vendors to virtualization technology providers to smaller and newer players developing virtual network services or offering hardware accessories and processors.

So I would separate this into 2 broad areas: (a) Physical Devices supporting SDN technologies and
(b) Virtual Network Appliances and services & associated virtualization technologies.

In the first category, a lot of well established and deep-rooted traditional networking players belong along with some of the newer players offering innovative products and solutions.

• The usual suspects such as Cisco, Juniper, Brocade, HP, Dell, Arista, Extreme all belong here. More so, because most of them have a number of their physical devices already enabled to understand Openflow and do VXLAN tunnel termination. Some notable acquisitions in this space are Brocade acquiring Foundry to beef up its networking portfolio, Dell Acquiring Force10 to expand its data center offerings, HP acquiring 3Com to foray into the networking space, IBM acquiring Blade to get into data center switching, just to name a few.
• Plexxi, which came out of stealth mode very recently and is a likely acquisition target, offers physical switches which are SDN-enabled. It also has controller and orchestrator components, and a set of APIs for applications. Their value proposition is to flatten out the network and so arguably could even belong to the Overlay Network Layer, but since they offer a physical switch, I am placing it here.
• Pica8 offers a number of 'Open Switches', which have OpenFlow and Open vSwitch integrated in them. 
• Vello Systems, which also is a very likely acquisition target, develops high-speed optical networking systems to reduce latency between data centers for latency aware services. 

In the latter category there are a number of players and acquisitions that have happened:

• Vyatta was acquired by Brocade in Dec 2012 for. It creates virtual routing and switching products
• Linerate was acquired by F5 in Feb 2013. It creates virtual network services.
• Embrane's Heleos product suite offers overlay-aware virtual network services such as load balancers and firewalls. It is likely to be a hot acquisition target.
• Insiemi, Cisco's $750 mm spin-in, is still in stealth mode, but speculations are that they are creating hardware (probably storage related) and some virtualized network services, all geared to fuel Cisco's SDN agenda. According to the rumors they are also building a Controller.
• XSigo systems was acquired by Oracle in 2012. It creates I/O virtualization software and hardware

Other Related M&A

There are other M&A activities that have happened recently. Although these are not directly related to SDN, but there is an SDN play here. Specially these are companies that started off building a product, but are being re-purposed for SDN.

• Cisco's $1.2 billion acquisition of Meraki is not clearly an SDN play. See the article that talks about the acquisition and its link to SDN here

Going Forward ...

2012 was the year of SDN. It witnessed good amount of $ flow in the SDN space - not so much from a revenue perspective, but more from VC's investing in SDN companies, and from bigger networking giants gobbling up smaller ones. Several start-ups came out of their stealth mode and talked openly about their products, and in no time were snapped up by bigger players. And the biggest of them, the Nicira acquisition, garnered a lot of attention from VCs, large companies, and entrepreneurs alike. A series of M&A activities ensued.

In my opinion, this M&A trend is going to continue for a while, as more innovation occurs in the different parts of the SDN layers. The application layer will see a lot of innovation, but that will happen once SDN is more widely adopted and some standardization of the northbound APIs are in place. The Orchestration layer will mainly see existing server provisioning systems and network management systems being repurposed to do network orchestration. The Controller and the Overlay Layer are again areas where there will be both M&A and innovation. The Underlay layer will see less of SDN-only innovation and more of generic hardware innovation and companies will enable their physical devices to understand SDN overlay protocols and interact with different controllers. As is quite evident, over the next few years, most of the SDN revenues will be from the Network layer and less so from the Controller and Application layers, but as SDN matures we will see reversal in that.

One thing that you will find interesting is that before SDN, networking industry was vertically integrated - as far as the different layers are concerned. Although there was separation between the control and the data plane, you could not for example have Cisco's control plane talk with Juniper's or Brocade's or HP's Data plane. Truth is, with SDN, the main players are also moving towards a similar vertically integrated model, which means that incumbents are all trying to have presence in all the layers of SDN, mainly through acquisitions, but also through organic growth. The direct upshot of this vertical integration is that you should expect to see a lot of M&A activities in the different layers of SDN - both from the incumbents and from new players (like VMware or Oracle) wanting to diversify into the space. We've already seen several such acquisitions, Juniper-Contrail, Cisco-vCider, VMware-Nicira, Citrix-Cloud.com, and we will definitely see more of this over the next few years. So what I am trying to say is that ..... before another VMware decides to shell out another billion to strengthen another part of their SDN architecture, why not go and build another Nicira. "Where do I get the funding ?" you ask...Well, I think it has so much attention from VCs that capital has already started 'flowing openly' or should I say 'open flow-ing' in this space...