Article Written By Sasa Nijemcevic, Vice President and General Manager, Network and Service Management Business Unit at Nokia

With over-the-top (OTT) applications and cloud-based services now the norm for both consumers and enterprises, expectations are rising for a perfect online experience. In an age where time demands are so high, "slow" has become the new "down", and network providers often find themselves at the receiving end of complaints about the performance of today's OTT applications. Dissatisfaction leads to churn, with subscribers quickly switching network providers in search of a better experience. Cloud service providers are also not immune with degraded application performance negatively impacting both their public image and revenue streams.

This issue has bounced between cloud and network providers - without satisfactory resolution - for some time.  Some of the largest web content providers have begun investing in their own networks to gain more control over customer experience. But they too have run into problems as multiple services began to vie for the same bandwidth, and because they have had to rely on cross-provider hybrid networks to reach all their customers.  At the root of the problem, for both cloud and network providers, is lack of the visibility and control necessary to identify and resolve OTT application impacting network issues quickly and cost-effectively.

Providers today have petabytes of data at their fingertips. However, this data is collected in silos across multiple systems, requiring providers to manually combine and correlate billions of data points amassed, organize them into coherent reports to try and sniff out issues, and into manual network reconfiguration plans to try and resolve them. The result is a painfully slow and costly guessing game that rarely resolves problems quickly enough to keep customers happy and costs in line. The solution, for both cloud and network providers, is to make their networks far more responsive to the needs of the OTT applications and services that flow through them.

Insight-driven control using software defined networks (SDN) and analytics is one key method being adopted today to solve this issue.  Many network providers worldwide that offer enterprise, wholesale and residential broadband services over their wide-area networks (WANs) are currently evaluating this technology within their current IP/MPLS and optical networks as well as for new cloud orchestrated NFV and SD-WAN service delivery architectures.  It works by correlating cloud and network data in real time to get an understanding of how cloud applications flow to and through networks, and leveraging this data to drive the underlying WAN so that it can dynamically meet fluctuating demand patterns and automate network service operations from provisioning, to optimization to assurance.

There are multiple ways to solve the OTT application experience problem with insight-driven SDN control. One way is in recognizing that OTT applications transiting the Internet are entering a high-risk area where their network provider does not have end-to-end control, and that in taking policy driven action to avoid the internet for high revenue potential applications would become the preferable option where available. The latter can be achieved by partnering with their network provider through building out dedicated networks to interconnect OTT application partners and using SDN control to intelligently steer applications.  In this way, application flows with higher bandwidth needs or latency sensitivity can be steered onto the best serving network resources - with a path through the network for which SDN analytics has determined will meet the specific needs of the application all the way to the data center. In the case of multiple data centers, insight gained from analytics will be used to select the path to the data center that is most optimal for meeting the application requirements. 

This all sounds great. But one must consider building out the network in this way can take months and may not meet the time-to-market required for new OTT application release launches.  And for many emerging OTT application providers this "build-out" option will not be cost effective enough to be viable.  So ultimately, in many cases, it will be unavoidable that the service will have to cross the internet at some point. However, even in such cases where OTT applications need to run over the Internet, there is still a choice to provide more insightful control using SDN. SDN methods like traffic flow steering have much appeal because they have less impact on the overall network, and often become the first choice to implement before dynamic network re-engineering.

Often times performance issues can be proactively avoided simply by dynamically selecting the peering point into the Internet that best meets a specific application's needs. Here's how this works. Through streaming telemetry based analytics, links to peering points with partnering interconnecting internet service providers, content delivery networks (CDNs) and OTT content provider networks can be monitored in near real-time.  When congestion or performance thresholds are crossed on specific network links, a centralized SDN controller can take policy-defined actions by leveraging an understanding of services and the flows that ride on these specific links.  For example, actions could be to perform an optimization by steering traffic onto an existing path, or even creating a new more controlled path (say by using segment routing technology).  This can be done on a per-service level, or even at a per-application level when there is integrated multi-dimensional analytics needed to provide the application-awareness that identifies and classifies individual flows.  Multi-dimensional analytics can bring visibility into the type of applications transiting into the WAN to enable more fine-grain application-specific threshold triggers for SDN control.

For example, with insight-driven SDN control leveraging multi-dimensional analytics, operators gain the application-level insight needed to assure video or gaming experience by selecting more optimal peering partners based on the health, utilization levels, and performance KPIs of the link to the destination internet service provider. (See figure below.

Figure 1. Insight-driven SDN control use-case for OTT applications and business internet services.

And in addition, all traffic flows entering the provider WAN can also globally benefit from insight-driven network optimization through a centralized SDN controller for automated quality of service classification as well as automated re-direction onto network paths that best meet service requirements.  Using SDN standards for a path computation engine (PCE), the network operator can even go as far as triggering the SDN controller to control the creation of new network paths that better meet SLAs.  The rationale for enabling the ability to dynamically create new paths in many cases is to have more path diversity to differ path placement decisions for network services compared to typical constrained shortest path first (CSPF) path computation based on distributed routing protocol-based traffic engineering data or reserved bandwidth, or especially offline software injected traffic engineered explicit routing options that are being used today.  These approaches are prevalent in routed networks today, yet is failing to meet the needs of OTT applications because of a tendency to select common paths that are most likely to be hit hardest at peak traffic periods.  This implicitly leads to the creation of network bottlenecks that cause application crippling network congestion.  It also does not maximize the full network investment because of over-provisioned bandwidth reservation where traffic does not get well distributed, leaving parts of the network under-utilized even at peak demand.

In recent years, advanced SDN approaches and algorithms for self-tuned adaptive routing (STAR) have been introduced on standards based SDN path computation engines (PCEs) to address this problem.  For example, the Bell Labs STAR algorithm [1] implemented within the PCE from the Nokia Network Services Platform (NSP) has been proven to yield better economics than the traditional approach of using constraint shortest path calculation based traffic engineering. In the case of the Nokia NSP, its PCE applies advanced control algorithms when performing path computations that leverage the Bell Labs STAR and take a rich feed of streaming telemetry data (e.g. interface statistics, link latency/jitter/packet drops) into account along with per-flow or per-service key performance indicators.

For all the cases discussed, whether by providing OTT providers with dedicated access to their data centers, or steering traffic to the best peering points into other internet service provider, insight-driven SDN control can help operators to automate and optimize for improving OTT application experience by ensuring the best network paths and exit points are continually being used to proactively avoid network congestion and meet service requirements at an application granularity.

Insight is paramount to driving the most intelligent automated networking and SDN control.  And as web scale service provider networking evolves to web scale deployments, insight-driven SDN control with integrated multi-dimensional analytics will be required to provide the operational agility and automation needed. This will enable the best quality of experience for the latest dynamic and bandwidth sensitive cloud applications that run OTT, whether virtualized in data centers, transiting content delivery networks, or transporting over SD-WAN.

With insight-driven SDN control, the network is continuously monitored with rich streaming telemetry (for example, on link bandwidth utilization, latency, jitter, etc.) that gives the needed visibility into the network state and resource health of the underlying provider network.  And with this insight-driven SDN solution providing the scale to handle the highest volumes in rapidly changing network demand patterns, providers can now proactively mitigate the chances of the network congestion that cause degraded performance so that OTT applications can continue to perform well.  The end result is improved OTT application experience with insight-driven network control. 

[1] "Computing a path to more profits - The benefits of a centralized Path Computation Element using Bell Labs Self-Tuned Adaptive Routing", Nokia Bell Labs,  https://resources.nokia.com/asset/186905

##

About the Author 

Sasa Nijemcevic is currently the Vice President and General Manager, Network and Service Management Business Unit at Nokia, drawing on 20 years of experience in the network management domain. Previously, Sasa has led R&D teams for IP/MPLS network management, wireless network management, and fixed/wireless policy management as well as Subscriber Data Management and Messaging products. He was instrumental in driving the success of Alcatel-Lucent's network management portfolio across the IP, Optical, microwave and wireless domains. Sasa has a Master's degree in Telecommunications and Computer Science from the University of Zagreb.