This Industry Viewpoint was authored by Priya Natarajan, Senior Director of Software Marketing, Ciena
Forecasting is inherently a challenge, but it’s absolutely critical in today’s business environment. For example, complicated sales funnels can shroud the view of how much revenue an organization will have to work with in the future, making sales forecasting tools like those available through Salesforce not only useful, but essential. In addition, inventory management tools can alleviate potential overstocks or shortages and financial forecasting solutions can help businesses weigh the impact of different options when making financial decisions. In fact, according to IDC, the market for Advanced and Predictive Analytics (APA), which serves as the most advanced method of forecasting, was $2.2 billion in 2013 and is predicted to grow to $3.4 billion in 2018 as more emphasis is put on forecasting in business.
Network activity forecasting is increasingly important for managing capacity fluctuation
And yet for all of the advanced predictive methods available today, forecasting when it comes to the network’s activity and demands can feel about as reliable as reading tea leaves. Just consider the industry analyst forecasts for the software-defined networking (SDN) market in 2018, which range from $3.5 billion to $35 billion. That’s quite a broad range.
For Web-scale network providers with software-based business models, forecasting network traffic and expected demands takes on a new importance entirely. As opposed to when a bad forecast from an inventory management tool causes a shop to run out of a particular product, when a network’s demands exceed current capacity, it can bring down the whole service or software for everyone—even the folks that got there before the overload.
And sometimes the need for capacity can come suddenly and unexpectedly. Take for example the issues Sling TV had when streaming the NCAA Final Four this spring—a huge influx of subscribers trying to stream the game at the same time crashed the system, and left many viewers staring at “error” messages on the screen. When Netflix releases a major hit, say, a new season of “House of Cards” or “Orange is the New Black,” viewer demand for bandwidth spikes. While rare, Netflix service unavailability is not unheard of when public cloud provider Amazon Web Services (AWS) has an outage.
Forecasting capacity without over-exerting resources is a delicate balance
It would be great if we could simply provision lots of network bandwidth to prepare for demand fluctuations and spikes, but the reality is that this goes against constraints of money and time. Over-provisioning the network is a waste of resources that would be put to better use elsewhere, which is why forecasting network activity has been relied upon so heavily. But forecasting is an art as much as it is a science, and is difficult to do effectively.
Network functions virtualization (NFV) offers some relief to service providers. While providers still have to forecast with commercial off-the-shelf (COTS) servers, NFV offers greater flexibility with respect to the services themselves, which can be built up or torn down as needed. With a pay-as-you-earn model, providers do not have to forecast the number of instances needed for a virtual firewall, for example. Instead, they can get the exact number of instances they need, when needed. Contrast this with the traditional hardware model that often has 30 to 80 percent spare capacity, ready to go but draining resources even when not in use, and the savings add up quickly. Since the COTS hardware can be used to host a variety of virtual network functions (VNFs), the savings are furthered by not requiring a new hardware install for every added function. With the right SDN application, it is possible to dip into shared pool of resources when experiencing excess demand.
Forecasting is here to stay…make it more effective and easier with SDN and NFV
While it’s not possible to entirely get away from forecasting, it is possible to reduce the amount of guesswork that still accompanies this effort and free up resources to allocate to other parts of the value chain. Software enabling network programmability lays the groundwork for an agile architecture that can deploy new functionality and scale up or down capacity in short order. Both SDN and NFV offer sorely-needed flexibility to network providers, and can help prepare them for unexpected spikes that even their best-effort forecasts aren’t able to account for.