Published on NETSCOUT blog
Wireless networks require a huge financial investment and effort to deploy and operate. As the number of use cases, the traffic load, and the adoption of real-time applications increase, the end-to-end network infrastructure will become even more expensive and more complex to manage. To extract as much value as possible from existing networks, service providers have to actively optimize them, not just to increase capacity or speed or to reduce latency, but also – and more crucially – to improve the overall performance as perceived by the subscriber and to manage traffic based on application requirements.
Optimization in today’s wireless networks is still very limited. Service providers have little flexibility in the allocation of network resources, and the best they can do is to push through as much traffic as their networks allow, circumventing any bottlenecks as traffic goes from the core to the devices. Fixed resource allocation (i.e., of functions tied to hardware elements) creates inherent inefficiencies, because each hardware element has to meet peak traffic requirements, operating at less than full utilization the rest of the time.
With virtualization, system operators can avoid this problem by optimizing the allocation of network resources on the basis of application and service requirements, and traffic demand. This flexibility gives service providers more value from their networks in terms of subscriber experience, return on investment, and ability to roll out and support new services.
With NFV and SDN, service providers can dynamically optimize resource allocation as a function of the traffic load. When they add automation, they can move a step further and fully realize the benefits from agility and flexibility. They can fine tune the networks as capacity and latency requirements, distribution of applications and services, network conditions, subscriber density and use patterns change over time. With this fine-tuning, performance in a virtualized network will exceed that of a legacy network with comparable processing, storage, and radio capabilities. Service providers still need to plan networks to meet end-to-end peak capacity, but they no longer need to plan for peak for each network element. As a result, a network expansion may require less equipment to meet the same performance requirements, and this translates into cost savings.
Yet there is a price to pay. Non-virtualized, legacy networks are static, and their performance is relatively easy to track because they are predictable and stable. Service providers have decades of experience in planning, managing and monitoring these networks – and in identifying and resolving issues as they arise. Virtualized networks are dynamic and can adapt to network conditions and traffic requirements, but they are more complex. Service providers have to learn to manage and monitor their networks in a new way if they want to get the benefits that virtualization promises. If they continue to manage virtualized networks as if they do with their static, legacy networks, the gains from virtualization are limited.
The most fundamental challenge is for service providers to get to know their networks – how they work, how they fail, how they can get fixed, how they can improve. On the surface, service providers have all the information they need in front of them: it is all in their networks. But the amount of data they get from their networks is dazzlingly large, and they need to sift through it to identify what data is relevant and will give them a true knowledge of network conditions, not just a collection of KPI metrics.
To gather the knowledge they need and leverage it to optimize their networks, service providers need to change the way they run their networks, concurrently along multiple dimensions.
A user-centric perspective. The goal of service providers is to provide the best experience to their subscribers and, increasingly, to their IoT clients. To achieve that, they need end-to-end visibility in their networks. Network performance should be assessed from the subscribers’ viewpoint: are they getting the performance they expect for the applications and services they use? Here, it is the perceived performance that matters most – what the user sees, not what goes on inside the network. These two aspects are related, but service providers that track performance only inside the network cannot capture users’ perceived experience. The network may work perfectly well – good KPIs, no network disruption – and yet the user experience may not be good for some applications, or the performance for some services may not meet the requirements. Downloading a big attachment may work well, but voice call quality might be bad.
Service-based granularity. Monitoring performance at the network level (e.g., throughput, latency, dropped calls) or at a function level is still necessary, but that has to be put into the context of how it contributes to the user experience for specific services and applications, or how it relates to variable performance requirements for IoT applications. For instance, the expected performance may be different for conversational video, video streaming, and video downloads. Video content is served in all three cases, but the latency requirements, for instance, are very different: video downloads are not affected by high latency in the same way conversational video is.
End-to-end monitoring and service assurance. To adopt a user-centric perspective, service providers have to look at the end-to-end network and drill down to different network areas, at different depths, for root-cause analysis and issue resolution. In complex dynamic networks, multiple and variable interactions shape the perceived user performance for each service used. Monitoring and service assurance have to relate the end-to-end, high-level, experience as perceived by the user, to the performance of individual network element or functions and at their interactions across the network. Going back to the previous examples, a service provider needs to know if the voice call quality or latency for conversational video is negatively affected by other traffic.
Real-time network management and resource allocation. Because virtualized networks are dynamic, service providers have to run them, optimize them and fix them in real time. The time resolution will vary depending on the task and the service provider’s capabilities and strategic choices, but the move to real time is crucial for reaping the benefits of virtualization’s flexibility. Real-time resource allocation makes it possible to increase network utilization, and this increases cost and performance efficiency. The causes of bad voice call quality or high latency for video will vary depending on network conditions, traffic composition, and service requirements, and the remedies will vary accordingly.
Closed-loop automation. Managing complexity is not easy. All the changes required by dynamic networks create a massive increase in the amount of data the scope of data analysis, and the number of actions a service provider has to deal with. Closed-loop automation is necessary to manage this new workload. Automation is much more than an expedient way to relieve staff of repetitive tasks or reduce labor costs. Closed-loop automation gives service providers the knowledge they need about their virtualized networks at multiple granularity levels. It enables them to use this knowledge to troubleshoot and optimize their networks – and to keep learning and honing their ability to improve network performance. Network virtualization both enables and requires automation to be an incremental and ongoing process, in which learning and optimization continue through time and strengthen each other.
Read more on NETSCOUT NFV Smarter solution
Read the companion interviews and blogs: