  | Issue: | Europe II 2014 |
| Article no.: | 16 | |
| Topic: | White box switches in the new data centre | |
| Author: | Steve Garrison & Seamus Crehan | |
| Title: | VP, Marketing, Pica8/ & President & Founder Crehan Research Inc. | |
| Organisation: | Pica8 | |
| PDF size: | 235KB |
About author
Steve Garrison is Vice President of Marketing of Pica8. Steve Garrison leads go-to-market and brand development strategies at Pica8, and is a networking systems veteran with nearly twenty years of technology marketing experience. Steve has held global marketing positions at public and venture-backed companies, including Infoblox, Force10 Networks (acquired by Dell, Inc.) and Riverstone Networks (acquired by Alcatel-Lucent).
He holds an MS in Materials Science Engineering from MIT, and a BS in Ceramic Science and Physics from Alfred University.
____________________________________________________________________
Seamus Crehan is President and founder of Crehan Research Inc. which specializes in statistics, analysis and insights for data centre switch, server, and storage networking. He is recognized as a leading data centre networking market expert and has made presentations and moderated at numerous conferences. Seamus has partnered with both big and small companies to help address various challenges and deliver quality research and insights. These companies include system and component vendors in the Switch, Server and Storage markets, in addition to financial institutions whose coverage and research includes these sectors.
Prior to Crehan Research Inc., Seamus tracked the Switch and Adapter markets at Dell’Oro Group for ten consecutive years, where he last served as Vice President.
Seamus has a degree in economics from Stanford University and an MBA from Santa Clara University.
Article abstract
White box (non-branded) switches are becoming popular, just as off-the-shelf servers have replaced proprietary PCs. Key attribute of white box switches is the common OSs that is able to plug in to OpenStack or to open-source OpenFlow controllers. Orchestration is achieved by SDN interfaces in the underlay, on a white box or legacy hardware, or by NFV server-based routing in the overlay (or in tunnels). While the Cloud segment is showing rapid growth, the traditional enterprise data centre market accounts for much greater switch volumes and revenues. The transition from general purpose switches to switches specially built for the data centre also goes against white boxes, so the adoption of them in the Enterprise is slower than in the Cloud segment.
Full Article
White box switches are getting more attention in next-gen data centre deployments, with a number of software-defined networking (SDN) startups offering solutions that include them. Enterprises are wondering how white box switches will impact their data centre plans. In this article, we’ll look at white box switches and see how they could potentially impact the market for data centre infrastructure.
What are white box switches?
White box has been a term people used to describe the non-branded manufacturers that built PCs. Many of these ODMs (original design manufacturers) ramped up production of white box servers, and now many of them are producing white box switches. White box switches look just like any other switch and are assembled by the same companies that build white box servers: Accton, Celestica, and Quanta Computer all make white box switches.
The software side of white box switching
White box switches are useless without software. Every switch needs an operating system (OS). The OS needs to seamlessly integrate with existing L2/L3 topologies and support a basic set of features. Beyond this, there should be new capabilities delivered as a result of opening up the network switch.
White box switches first and foremost need to have hardware-agnostic network OSs. This enables an abstraction layer on the bare metal switch, just as Linux/Windows does on the server side. White boxes are an ideal metal for Linux based OSs. Linux offers many great benefits (including open/free tools like GCC, a native environment for Python, and the ability to compile your own app on board) that form the best foundation for leveraging the benefits of white box switching.
How do you put the OS on the metal? Some vendors are selling a complete solution with the OS on the white box, and some are setting up distributors to provide the metal and you buy the OS direct from the software vendor. Both approaches have merit depending on the scale of the deployment and the desire to have a single or dual source of accountability.
Beyond the operating system, white box switches are more valuable if they interact with SDN controllers. Therefore, the best switch OS products will have interfaces to different controllers. Most people think of OpenFlow as a protocol when we talk about controllers, but really Centre Orchestration from VMware is a controller: something has to “control” the VM lifecycle. So, we think a key attribute of white box switch OSs is to be able to plug in to OpenStack, or to open-source OpenFlow controllers like Ryu (http://osrg.github.io/ryu/), Floodlight, or most recently Open Daylight.
SDN, NFV, and NV
How does the OS on a white box switch participate in orchestration, such as being able to support overlays or network function virtualization (NFV)? Two different approaches appear to be emerging in addressing these pain points. The under layer or physical network infrastructure approach is focused solely on the network fabric itself. You might deliver your solution on a white box or a legacy vendor switch with your own hardware design. Nonetheless, you care about routing/switching/management and provisioning the network itself. You might support OpenFlow, and as noted, these solutions are delivered on a switch. The ASIC (application-specific integrated circuit) in the switch is an integral part of the story.
The overlay/virtual network infrastructure approach is focused on building tunnels (also called overlays) and that aspect is called Network Virtualization (NV). Building virtual network function capability into those tunnels is called NFV as mentioned earlier. For example, you may want to connect two isolated domains without changing the underlay and possibly spin up a firewall on that tunnel. These solutions are delivered on an x86 server platform. There are no ASICs and the rationale is that you can scale here through all of the multicore CPUs in a server.
Many prospective customers are starting to understand the difference between under-layer and over-layer segmentation. In some cases, the under layer focus is what the market is calling SDN. Anything running on a server is either NV or NFV.
The market for white box switches
So what does the market look like for white box switches? We can start by looking at the overall data centre Ethernet switch market. Then, within this, we can look at the cloud (web 2.0, portals and hyperscale) and enterprise segments separately since these are on different trajectories and technology adoption rates.
The data centre Ethernet switch market has seen tremendous growth and investment over the past four years, despite an overall challenged and constrained IT spending environment (Figure 1). This revenue growth is even more impressive given the shift to fixed/top-of-rack switch architectures at the expense of more expensive modular/chassis data centre switches. Fixed/top-of-rack deployments far outpaced the overall data centre switch market with revenues more than tripling.
Within the data centre Ethernet switch market, the cloud segment has seen exponential increases as new companies, business models and services are enabled by massive data centre networking infrastructures. Crehan Research estimates that in 2012 the total cloud segment accounted for about four million data centre Ethernet switch ports – including both OEM switch brands such as Arista, Cisco, Dell, IBM, Juniper and others as well as white box switches from companies such as Accton and Quanta Computer, as noted before.
Although a few very large cloud data centre companies have deployed white box data centre switches, we think the majority currently uses OEM branded data centre switches. On large deployments, the price difference between the white box and the discounted OEM branded switch may not offer a compelling enough ROI to change to a white box model while on smaller deployments there may not be enough scale to make the white box economics work. However, there continues to be more interest, offerings and business models around white boxes.
Crehan Research expects that the cloud market will see about a 25% shipment CAGR over the next five years, growing from under four million ports in 2012 to over twelve million ports in 2017. Additionally, this segment will likely see at least a gradual increase in white box deployments. Combining these two factors results in a 32% shipment CAGR and a forecast close to five million data centre Ethernet white box switch ports deployed in the cloud market segment by 2017 (Figure 2).
As we can see in Figure 2, while the cloud segment has seen a lot of growth (and has got a lot of attention), the enterprise or traditional data centre switch market is still by far the largest driver of data centre Ethernet switch volumes and, even more so, revenues.
In addition, the enterprise data centre switch market has been an area of change and innovation. Over the past decade or so, enterprise data centres have moved from distributed to centralized and consolidated; physical to physical plus virtual; and siloed to converged. At the same time these data centres have had to handle a step-function shift in the demand curve for networking traffic and its storage, in conjunction with multiple new client devices (such as smart phones and tablets). All of this while working with legacy infrastructures, since they do not have the luxury of starting over again with a clean slate.
In conjunction with all these changes in enterprise data centre switching, we are seeing the transition from general purpose switches deployed in data centres to switches specially built for the data centre. This is shown by Cisco’s transition from Catalyst (general purpose switch including data centre) to Nexus (switch specifically built for the data centre) and the rise of data centre switch specialist Arista Networks.
As Figure 2 shows, even with a much slower growth rate than the cloud, we still expect the enterprise to remain as the majority of data centre Ethernet switch market through 2017. Within this market we expect to see much slower adoption of white box switches because of factors such as more diverse environments, longer and closer ties to existing switch vendors, end-to-end or full solution purchases, and vendor-provided value added services such as technical support. Furthermore, with an average data centre switch lifecycle in the enterprise in excess of four years; we need to keep in mind that a migration to new deployment models will be gradual for this segment as a whole.
Combining both the enterprise and cloud white box adoption curves results in Crehan’s prediction of three million ports in 2014 or 7% of overall shipments. By 2017, we forecast that this will have increased to over eight million ports or 15% of total data centre switch port shipments. So while we believe adoption of white box switches will be gradual, they can nonetheless be an important part of the data centre landscape over the coming years. SDN will flourish with open systems, and white box switches can be an important part of that growth.
