- Unpredictable network demands dictate the need for dramatic change
- A cloud technology toolkit can guide infrastructure transformation
- The future requires an optimized, programmable network
As cable and telco operators plan for a connected future, they start to embrace cloud concepts such as SDN and NFV as part of their network optimization strategies. New business models and applications provide opportunities in this future, but how do operators ensure their position in new value chains? Like it or not, as customer demands increase and technology changes, we need to dramatically change the network infrastructure.
NEW DEMANDS ON THE NETWORK
Market trends set out the challenge that we have in building networks. Just in terms of devices and people connected, traffic growth has been phenomenal.
In recent years, traffic growth has been largely driven by video, which places an increasing load on the network infrastructure. And video is changing. Catch-up TV, for example, shifts video from broadcast to unicast and delivers individual streams to subscribers -- which increases capacity requirements on the network infrastructure. How do we build networks that can scale to meet these new demands?
The Internet of Things
Besides more users coming on to the network, many more devices are driving up the amount of connectivity that’s required. And these new devices are no longer I/O bound.
Humans deliver content to humans. Humans are I/O bound. But machines are not I/O bound. So with IoT and machine-to-machine communications, a new class of devices is collecting/sensing information that in theory can just continue to drive capacity forever.
On the other side of the network are all the servers and services that you want to reach. Many of the forward looking applications today are cloud based applications, and we need to access these compute resources that are in the cloud infrastructure. The aforementioned increase in capacity is amplified further by cloud computing. We are seeing a huge amount of traffic currently going between servers across datacenters. And in the middle of all of this, ultimately, is the network.
We need the highest performance, lowest cost per bit, highest capable network infrastructure to connect all this together. The challenge is: How do we get there?
We’ve previously been, in our industry, able to track network demands since they were largely driven by human behaviors. We had call models.
You knew Mother’s Day was going to be the busiest day of the year, and could plan for that. You knew people were going to make phone calls at Christmas time. Similarly, you can track what video people are watching and at what times of day. Humans follow patterns. And if we see patterns, we can reflect that in the network.
But now traffic is starting to shift from human driven workloads – entertainment, information, us doing our daily jobs – to machine driven workloads. Servers are running various jobs, and machines are not very predictable.
If you couple that with the aforementioned potentially limitless capacity, the size of the cloud infrastructure gets bigger. Machines get increasingly powerful with more cores, faster processors, etc. So we are dealing with a potentially much greater demand that’s absolutely NOT predictable.
How do we build a network infrastructure that can handle that?
Ultimately, the network itself has to become programmable. We can no longer do the capacity planning cycles that project when we’ll need to add capacity. We need to get the network itself into a position where it responds to those applications. And as those workloads are changing and shifting, the network and the underlying fabric in the connectivity mesh is able to adapt to those workloads over time.
We will have a new type of IP network. The end goal is to have a network that:
- Is re-programmable
- Can be re-repartitioned
- We can allocate bandwidth to and apply to workloads
And at the same time, we need to continue to scale capacity, we have to continually offer much higher access speeds, and we need to deliver new types of services. All this has to come together across a common infrastructure.
We also have to think how we manage and operate the network. As the network becomes more programmable, the systems we use to drive the network such as OSS and BSS need to become more agile.
NETWORK OPTIMIZATION TOOLKIT
You have seen a lot of the terms shown in Figure 3 – and together they form our toolkit for network evolution.
The end goal is a common IP fabric that is programmable, easy to scale, and that drives down cost per bit. And the technologies all must interoperate in some way. Concepts and technologies in the mix include:
- Software-defined networking (SDN)
- Network function virtualization (NFV)
- Segment routing
- Open source
- Tools such as Data Plane Development Kit (DPDK)
- White boxes
SDN is a concept, not a technology. It’s the idea that we can make a programmable network. We can begin to simplify and flatten the network by pulling functionality that used to be in the network - control plane signaling, signaling of end to end connections, managing bandwidth – and move it into software that runs in a separate layer that controls the network resources themselves.
Virtualization is another tool that we talk about. NFV with SDN is yin and yang. NFV is a technology, but still a bit of a concept. What functions will we virtualize? The entire function? Or part of it, such as the control plane? We can virtualize that, run it on a server, and couple that to some actual physical boxes that do packet forwarding. So virtualization is really a tool to simplify provisioning and drive a lot of the automation.
In terms of flattened hierarchies and simplification, segment routing really helps by allowing node level control and path control through the network without overloading the network with signaling protocols. Making it simpler.
Converging the network implies collapsing layers, bringing IP technologies, optical technologies and access technologies together in the same systems. These can be single shelf systems, multi-shelf systems, or be managed together and integrated at the box level. We want to make the network simpler.
We need to look at lower cost technology. Things like “white boxes” really mean “we want to drive costs out of the network”. Things like virtual silicon also drive costs out of the network.
Together, our toolkit for network evolution is all of these technologies, approaches, and techniques combined to optimize the network infrastructure.
THE OPTIMIZED, PROGRAMMABLE NETWORK
Figure 4 represents all these concepts and tools graphically.
At the base level is the network infrastructure. We still need to move bits in physical boxes or servers running software. And we need to implement those services within some type of hardware.
Then we need a control layer, and this is really the SDN control layer as you see in the middle. A primary goal of the SDN layer is abstraction. Network elements will continue to become complex, but they’re simplified in the SDN layer. It’s running on some servers in the cloud.
It’s going to be a combination of both service and network understanding. The services represent the duty cycle and workloads that are coming onto the network. So if you don’t understand the services, there’s no way to map that capacity into the network. And of course you have to understand the network itself. What’s the current status? What’s the utilization? If we bring together the services and the actual network, we can achieve network optimization.
Our applications and OSS/BSS systems that will ultimately be simplified are on top.
The OSS for example is dealing with fairly high level constructs. We abstract the complexity away through the SDN layer -- in many cases all the way down to the box -- to create a much more programmatic and dynamic environment.
In our toolkit, concepts and technologies will work together to achieve network optimization. The result is a programmable, common IP fabric that’s easy to scale and drives down costs per bit -- and in this way responds to unpredictable future needs as we move into the cloud era.
Our next article will provide insights on how our customers think about virtualizing edge functions and applying SDN to B2B services.
Network Services Platform feature page
Software-defined networking solution page
NFV solution page
Segment routing and path computation element white paper
To contact the author or request additional information, please send an email to firstname.lastname@example.org.