With market forecasts ranging from $6bn to $12bn in 2019/20, and Gartner saying 30% of users will be managing their WAN through software in the next 3 years there is some understandable hype and attention around this term today.
So what is SD-WAN and should we care?
The answer to the second part is probably yes. Whether it is an Enterprise using software to manage their own Wan, or SPs using software to provide more flexible services to customers, new models will come to the fore.
With the early promise of ease of deployment, central manageability, reduced costs and faster speeds to service provision, who is not going to want some of this?
SDX Central defines SD-WAN as follows “The software-defined wide area network (SD-WAN) is a specific application of software-defined networking (SDN) technology applied to Wan connections, which are used to connect enterprise networks”, so clear as mud then. To be fair it is hard to get a straight answer as to exactly what SDN is, but if you add the word “technology” you are covered, making it possible to at least broadly outline some of the technologies involved.
Of course an SDN isn’t a “thing”. Not a “thing” you can buy. It can range from separation and centralisation of the control plane from the data plane in a pure sense, on a controller of sorts, (of which there are many flavours), controlled by software (a variety here, open source, closed, hybrid) to provide programmability of the network (several ways to do this), to the configuration of the network (APIs are the new CLI), and configuration management tools and orchestration engines (Salt, Chef, Puppet, Ansible).
Ultimately much of this is trying help the network keep up with the speed of change higher up the stack where some of these techniques have been used for a while (server, compute and storage worlds). As we know, the speed of change in networking is traditionally slow.
Much of the above relies on a variety of abstractions and overlays to attempt to streamline services or hide complexity.
It is tempting here to drift into a discussion around abstractions, scale, state, control, complexity, speed and intent at this point, but I’ll save that for another day when I have more energy.
Some see SDN as an architecture and SD-WAN as a product you can buy?…ok, maybe.
Or maybe more a solution-set and platform for vendors in this space (Anuta-networks, Talari, Cloudgenix, Viptela, Pertino/Cradlepoint, Velocloud, Glue/Cisco, to name a few.)
The point of all of this? To make networks more predictable, cheaper, quicker to react, more controllable, stable, service orientated, and accelerate time to market. The usual stuff then.
What I have seen lately is an increasing acceptance of flexibility and utility in the minds of network engineers. The idea of spinning up a service across the stack as you need it and tearing it down as soon as you don’t, or it becomes too costly to troubleshoot or maintain (it can be easier to spin up a new one than fix the existing). These ideas are seeping slowly into the networking world. I am not saying it is necessarily the right way, but certainly something I am seeing.
So the next question is, how would we enable good old-fashioned networking to take on such flexible new-age characteristics in the Wan?
One way is to use some of the techniques of SDI–Software Defined Infrastructure. Put simply, “Orchestration and Management software around Storage Compute and Network which Automates provisioning and configuration.”
Combine this with SD-WAN and Network Function Virtualisation and you can easily visualise developing a managed router service deployed with NFV to virtual CPEs (vCPE).
What is NFV? Well NFV decouples network functions, (NAT, DNS, intrusion detection, firewalling, load-balancing etc.) from proprietary hardware appliances so they can run in software.
Ok, now we’ve got that straight…let’s look at some implications for the Wan at a high level
What we are looking for is flexible appropriate network access for the right application at the right time.
One possibility is to consider a hybrid-Wan. By that I mean dynamically routing traffic over private and public links when it suits the applications e.g. MPLS and LTE/broadband/wireless respectively. This certainly looks like low-hanging fruit for SD-WAN.
Say you are using an SD-WAN type service connected to private MPLS but with broadband/3G-4G/Wifi as backup, and you can get extended visibility into these networks around reliability and performance. What some have seen is better performance and reliability in some SP environments over 3G/4G than their existing private expensive MPLS service, so is it beyond the pale to consider flipping the priority and trying public first for high priority apps?
If so, could we see a ramping down of the private MPLS circuit and replacement with DSL internet?
Then along comes 5G…
It is worth asking the question, is your next Wan a 5G network?
5G takes the idea of DSL internet a step further. 5G isn’t simply increased bandwidth. It will be seen operating in several spaces, from low bandwidth, short transaction IOT (machine to machine) chirp devices, to enhanced high bandwidth applications using some of the advances we have seen on the WiFi side in the last few years (MIMO etc.), and segmentation thereof. Of course the higher the bandwidth the lower the effective duty cycle for different applications – download your Netflix/Prime/Now/Youtube video quickly and get off my network! Freeing it up for other uses.
5G has the potential to offer improved indoor coverage, low power, large numbers of connections per cell, and machine to machine expansion.
For this Network Function Virtualisation (NFV) will play an important role with SD-WAN in a 5G network where lower latency will be key for IOT devices at the edge of the network. Figures being thrown around at the moment are 1ms latency with 5G as opposed to 50ms with 4G, a million connections per km squared, and 10Gbps throughput potentially. Maybe even real-time Cloud-RAN performance of micro rather than milliseconds? From this you can see how NFV and processing with standard hardware towards an intelligent edge / Fog Computing architectures (Openfog) will start to make sense.
Currently the ETSI Industry Specification Group for Network Functions Virtualization (ETSI ISG NFV) includes all the major players, Telefonica, Verizon, BT, Deutsche Telecom, AT&T, Orange etc. and has grown to include over 230 companies in the interest of trying to drive standard IT virtualisation technologies. Network functions on industry standard server hardware make it very easy to move locations and reduce the need to install new, expensive, proprietary hardware every time you introduce new services.
C-RAN – SD-WAN and NFV seem a natural fit.
Cloud-RAN (C-RAN or sometimes centralised RAN), and small cell architecture is based around centralisation and virtualisation. It is therefore easy to see how the above techniques will play an important role in 5G, e.g. 5G small cell deployments in the 30ghz band.
With 5G and C-RAN there will be a diverse range of use cases and requirements that Service Providers will need to be able to respond to quickly.
Cloud-RAN, with the separation of the Base Band unit (BBU) and the Remote Radio head (RRH) using fibre, 5G technologies, mmWAve (30Ghz to 300Ghz VHF/EHF) with CWDM/DWDM to extend baseband over long distance, will enable the centralisation of control into large scale centralised base band deployment. These technologies enable dynamic resource sharing, virtualisation, low latency, high bandwidth and reliable interconnect to a BBU pool. We will see more collaborative BBU technologies, and open platform real-time virtualisation technologies.
Different hierarchies will contain different RAN radios of different sizes using multiple data rates. Cloud-RAN will enable variable rate cells. How can these be provisioned quickly and cheaply?
How do you enable the various use cases created by retrofitting narrowband to 3g/4g? Narrowband, as its name suggests, essentially uses a narrow-band of frequency spectrum to provide discrete bands (from 20Mhz to 200khz wide) for lower data rate coverage (half or full-duplex) and now that 3GPP Narrowband IOT standardisation is complete, you will surely see a variety of use cases at different speeds for different applications especially with IOT.
Software Defined Infrastructure (SDI) certainly looks like a good fit here in providing more flexibility and automation to service provision at the edge.
So where does all of this leave the successful SPs of the future? Providing enhanced visibility across their services allowing Enterprises to make better decisions? Granularity and speed of service provision? Virtualised network functions, security? 5G capabilities?
Long term private Wan contracts should become a thing of the past with this new flexibility, and as the Cloud continues to prove popular as a service model, many customers will simply want “secure, flexible, and reliable access” regardless of how it is delivered..
As ever, it will be the ones who are seriously thinking about all of this now who will come out on top. With such rapid innovation and development, it is fine not to have the solution today, but we really do need to invest in the problem!
One final note, all these things tend to hype the way to the holy-grail of simplicity. Smarter, self-aware, self-optimizing, self-scaling, self-healing networking – who wouldn’t want that?
Some of the things covered will certainly help to make some of what we do today simpler, but with increased flexibility will come new services, and at scale this will always be complex. Networking is complex, we can abstract away some of it (hide it in an abstraction), but there will always be complexity.
The popular TV series “Heroes” had the tag-line “Save the cheerleader, save the world”. In the less dramatic world of modern networking maybe it should be “Understand your abstractions and interface points, save the world”