The Rise Of Multi-CDN & CDN Selection

For a long time, CDN selection and multi-CDN were in a chicken-and-egg situation – how could one emerge and grow without the other? To many OTT Operators, multi-CDN was not a priority because price and quality were good enough from their single supplier. Plus, a multi-CDN set-up brought new complexities such as managing multiple suppliers, inserting an extra layer of technology into the chain, and supporting multiple delivery points from the Origin.

But things changed as content delivery volumes increased. The CDN market began to see significant price competition. CDNs started to have failures and “good-enough performance” was no longer good enough for OTT Operators seeking strong business results from their growing investments.

Today, the standards expected from a streaming service are high, both from consumers and from media executives. Quality must be as close to broadcast-grade as possible, to deliver an excellent Quality of Experience (QoE) which leads to better viewer engagement and lower customer churn. Delivery cost must enable a streaming service to economically scale as the volume of content delivered grows exponentially and peak audiences expand towards the size of traditional prime-time linear TV audiences.

With these market-level changes, multi-CDN has taken off and solidified the demand for CDN selection solutions.

Initially, CDN selection was simple. The primary function was to allocate enough traffic to each CDN to ensure that committed traffic levels and expected price per unit of delivery were achieved. But as multi-CDN has matured, and the business importance of streaming has grown, CDN selection has become much more sophisticated.

Now, OTT Operators require CDN Selection solutions to deliver multiple business-critical benefits such as avoiding service outages and reducing churn, eliminating capacity bottlenecks, enabling a global streaming product strategy with consistent distribution platform implementation, and tailoring distribution methods to meet various budgets and performance needs.

Today’s Best Practice

From a technology perspective, there are three areas of focus to assure a high quality of streaming content delivery. First, the actual delivery platform itself – the encoder, packager, origin and CDN must all be set up for successful delivery. Second, the player – ensuring it works well with the wide range of devices. Third, the choice of profiles – encoding profiles must be handled well by the whole delivery chain from encoder to player.

CDN is the primary item in these focus areas where second-by-second dynamism exists. It is important to consider subjects like network performance variability, network scalability, network location, network type, network capacity, and potentially highly variable costs (figure 1). In other words, the way the whole network infrastructure performs after the Origin and before the Player can be impacted by many factors that are outside the control of the technical decisions made by the OTT Operator.

Therefore, today’s CDN Selection best practice is based on sophisticated and real-time decision-making logic to adapt quickly to changes in the network environment that affect stream performance.

Because different CDNs will have different answers to the questions in Figure 1, OTT Operators need to closely manage their relationships with each CDN provider. The focus on multi-CDN has led to OTT Operators working to obtain cost parity between CDNs, so that the CDN selection process can predominantly focus on performance, which is where the impactful day-to-day variability is observed.

Within the focus on performance, the subjects of CDN capacity location, capacity transparency, and capacity guarantees have become the latest hot topics. Simply “delivering 50PB in a month” is not enough. The delivery of those 50PB must meet exacting performance standards. It is like asking a courier to not just drop the package at the right address, but also ensure the customer receiving it is impressed with the condition of the package and the level of customer care and attention provided.

Leading OTT Operators today engage in dynamic CDN measurement and selection. This enables real-time decision-making at every moment of the day. CDN capacity utilisation (i.e., consumer demand) can vary easily depending on consumer behaviour and the content publication activities driven by all the businesses simultaneously utilising the CDN capacity. Capacity supply can also vary due to outages and planned maintenance, not only at the CDN level but also at the ISP level which can force large-scale stream re-routing. Regardless, the streaming show must go on and the streams must somehow be delivered well.

Dynamic CDN Measurement & Selection

Leading CDN Selection solutions enable OTT Operators to operate dynamically. They offer powerful configuration abilities that can apply, in real-time, all defined business rules at a very granular level and optimise stream performance across a range of KPIs that can be defined and prioritised by the OTT Operator.

The theoretical design of these solutions is relatively straightforward. The acid test is how they practically perform under real-life conditions, handling all required real-time computational processing. Although it may be possible in the CDN Selection system to drill down to device type and operating system type by country, ISP and ASN, and make delivery choices depending on live vs. VOD and the type of bitrate profile for different content, the reality is that making large-scale, real-time CDN selection choices at this level might be impractical. But as the world of streaming continues to evolve, these granular capabilities will become critical to success.

As shown in Figure 2, CDN selection is an initial task for each stream request, which becomes a continuous task as the stream is delivered to the consumer over time. Some CDN Selection solutions start with the premise that multi-CDN selection will first be driven by commercial rules that are fixed, such as minimum delivery commitments in a contract. This will set an initial business rule to deliver a certain quantity to one CDN which will primarily be managed by allocating a percentage of traffic to that CDN. This percentage-based solution, which depends on accurately forecasting total content delivery to allocate delivery correctly between CDNs, can lead to under- or over-utilisation of a particular CDN. Therefore some solutions have moved away from having percentage-allocation as a primary rule and instead they allow CDN Selection decisions to initiate with a wide range of factors, like capacity availability (e.g., I have 1 Tbps of capacity on my private CDN that must always be fully utilised), region (e.g., I deliver all my content in Germany, and I should deliver 100% of my content in the Frankfurt region to ISP “1” using CDN “A”), and content type (e.g., deliver 4K live content on CDN A, deliver all other content on any CDN).

The granularity of these business rules depends on how the content, network, and consumer are understood by the system. If the understanding is high-level - “the consumer is on ISP 1 in Country 1” (i.e., a simple DNS approach) - then CDN Selection simply passes responsibility for good delivery to the CDN. But if it is possible to understand the consumer’s location more precisely (e.g., through eDNS0 implementations) and profile (i.e., with client-side data), and it is possible to understand how the CDN is configured in terms of server locations, capacity availability, connectivity types with ISPs, etc., then CDN selection can be highly sophisticated, and the OTT Operator can have more control of the delivery process.

Once a CDN has been initially selected by these business rules, then the next step is to complete the selection based on performance. Using performance checks is a win-win for the OTT Operator and the CDN, because content will only be allocated for delivery on a CDN if it is performing well enough at that time.

Performance decisions need to be based on QoE and QoS, as measured by the client-side tools and the server-side tools. QoE takes priority in most solutions, which includes measures like rebuffering ratio, start-up time, start-up errors, and sustained bitrate. The CDN Selection tools must have granular, real-time data for all these metrics to support good decisions.

“In the NPAW Suite there are over 70 different quality KPIs and more than 90 different metadata elements associated with each playback,” according to NPAW’s CDN Balancer Business Unit Head, Luis Lopez Chousa. “Analysing this multi-dimensional data matrix in real-time can involve millions of calculations, which means OTT Operators can make pinpoint automated decisions about how to deliver their content given the environmental conditions at the time.”

Jose Jesus, Director of Product for Experience Insights Suite at Conviva, states “OTT Operators normally set up business policies to meet their CDN commitments while at the same time using each CDN when and where it performs best. It is important that algorithms optimise performance within business policy constraints, but if a CDN’s performance deviates too far from the baseline then it needs to be temporarily avoided. Typical metrics OTT Operators want to improve are connection induced rebuffering ratio, bitrate, technical video start-up failures, and technical video playback failures”.

Once a CDN has been selected Switching Modes ensure that the usage of each CDN is maximised for ongoing cost and performance. There are multiple switching modes as described in Figure 3.

Performance and cost results both depend on how the CDN Selector performs. Leading Technology providers are working on optimisations required by the leading OTT Operators, to push the performance of CDN selection to the next level.

Managing Public And Private CDNs

OTT Operators are increasingly investing in their own private CDN platforms, dedicated to their own content delivery. CDN Selection solutions may measure each platform in the same way but they should be able to allocate content to each CDN environment using different methods. For example, a private CDN should bring cost and performance benefits that need to be maximised. So, it should have the majority of traffic allocated to it unless there are problems. CDN Selection solutions that initially select using the percentage-based business rule are not ready to support this requirement.

Per Segment Delivery

A recent development in leading CDN Selection solutions is to enable CDN selection on a per video segment basis. A plug-in component at the player determines which CDN to retrieve each segment from based on a range of real-time measurements provided by the client-side analytics. Policies can be set that limit the ability to retrieve from a CDN that is not desirable for the OTT Operator, if there are performance or business constraints. This low level of granularity will provide OTT Operators even more methods in the future for optimising their content delivery.

NPAW’s Luis Lopez Chousa states “The per segment approach takes CDN selection capabilities to a new level that could transform streaming performance for the largest audiences. OTT Operators will be able to know that they are doing everything possible to extract the best performance from their delivery platforms, which is particularly important during the biggest events when the pressure to perform is at its highest.”

Reporting

Completing the CDN selection process is the reporting on decisions made and their results. Leading solutions allow for deep drill-down into session-level data which enables analysts to dissect performance and establish appropriate improvement plans. Real-time alerts of under-performing CDNs or content types can be provided and are connected back to the analytics reports. This feedback loop ensures continuous improvement is embedded in day-to-day CDN selection decisions.

Results

The OTT Operators taking this subject most seriously are seeing important performance improvements. Conviva’s Jose Jesus states “Our published reports over the last 7 or 8 years show that the industry has moved from looking at average rebuffering ratio, or the percentage of sessions with less than 2% rebuffering ratio, to measure how many sessions have zero rebuffering, as this represents a proper TV experience. Benchmark rebuffering ratio in Europe and North America, for example, has dropped from 1% to less than 0.1% over the last 7 years because best-in-class OTT Operators are continuously measuring performance and making improvements.”

Technology Developments

The general technology drive in CDN Selection is for more data, more real-time analytics, and more detail in decision-making. Additionally, some very specific areas are being researched and developed to support OTT Operators’ future needs.

• Internet control planes – to prevent risking an overload of a CDN, leading to performance degradation, customer dissatisfaction, and CDN switching, the CDN Selector must know when a CDN is full. This requires some form of control plane for CDN capacity decision-making. The following Streamer Requirements section explains this further.
• Deep Edge deployments – improving performance most often involves tuning how the players manage the video and working with CDNs to fine-tune capacity availability and locations. The placement of Edge Caching capacity nearer to consumers is a major trend that will ultimately help all OTT Operators to scale their businesses.
• Multiple delivery sources – the Per Segment Delivery approach previously mentioned can be extended to incorporate other delivery sources, like WebRTC or Peer to Peer. This could mean that specific segments could be downloaded from a non-CDN source, like a peer, potentially reducing costs and latency for the OTT Operator.

Streamer Requirements

Advanced OTT Operators, pushing the boundaries and requiring the very best solutions, describe three specific areas of requirement for their future CDN selection solutions.

First, business rule configuration is a must-have. Solutions that focus on technical selection criteria and do not include the ability to use finely tuned business rules are too simplistic. Fine tuning involves, for example, delivering lower value content on a lower cost / lower performance CDN, while higher value content should use the highest performing CDN. Given that CDNs can differ in each country in terms of features, performance, and price, it is important to use their services intelligently to drive best cost.

Second, the CDN Selector is the initial point of the stream delivery before streams spread out inside the CDNs. The CDN Selector is aware of exactly what type of device/consumer a stream is to be delivered to – for example, mobile vs. SmartTV, or a standard vs. premium customer. Therefore, it should be possible to manipulate the manifest at this point. With this functionality, the right number of users can be sent to specific CDNs with capabilities that best match the delivery requirement.

Third, data integration from 3rd party systems is critical. A prime example is the private CDN platform that OTT Operators can have very deep insight into, such as capacity availability by server, and the ability to control the amount of available capacity at each PoP. The requirement is to bring that data into the CDN Selection tools. Today, a lot of CDN Selection functionality requires installing an App plug-in which requires opening the App and inserting the data, and time and cost to integrate. It would be more efficient to bring data from the CDN into the CDN Selector that can be used to drive more granular capacity allocation. The basic rule of streaming performance at a CDN level is that the best performance is achieved when the right capacity is available. Pinpointing the right capacity in real-time would be a major benefit to the OTT Operators.

As Vincenzo Roggio, Head of Distribution at DAZN, states “In our large-scale live event streaming delivery, traffic is spiky. It starts, reaches a high peak, and then stops, many times per day. We need to know exactly how our CDN capacity is performing, and we need to know that during a 2-hour game we are getting optimal decision-making from our tools that help our CDNs to perform their best. We know this leads to the best QoE for our customers, which is business-critical for us. We have a multi-CDN environment and look for every way to continuously improve the performance and efficiency of our content delivery operations.”

Conclusion

CDN Selection has a big role to play in delivering broadcast-grade experiences for streaming viewers. It supports OTT Operators to optimise their CDN performance and cost, both of which are critical to their success as streaming grows.

The leading solutions are laser-focused on improving streaming performance through excellent use of data that delivers clear insights, actionable information, and a continuous improvement capability.

We can see that the streaming world is improving in every respect. Devices are better, encoding standards are sharper, networks are better. But, there is always the chance of encountering network congestion. To protect against this impacting stream performance or cost, analytics must be granular, and CDN selection must be dynamic.

In the end, whether a viewer receives their broadcast-grade streaming experience or not relies heavily on which CDN is selected.