Online video providers looking to ensure high-quality experiences for their customers are increasingly looking at using multiple CDNs, to ensure load-balancing, optimal route selection and reliability.
In Part I of our interview with Jon Alexander, senior director of product management at Level 3 Communications Inc. (NYSE: LVLT), we discussed how to prepare networks for a new age of streaming media, and the impact on intelligent network management. Alexander oversees Level 3's media suite of products including the global content delivery network (CDN) and Vyvx broadcast solution. (See Level 3's Alexander on Delivering 4K Video.)
Following that initial discussion, we had some more questions for Alexander on the benefits of using multiple CDNs, and how best to manage them.
Telco Transformation: Some content providers are using multiple CDNs to deliver their streaming services. Do you think this is a good approach, or are there significant logistical inefficiencies involved with managing multiple CDNs?
Jon Alexander: We support multi-CDN deployments and view it as a best practice for large-scale content delivery, where performance is important to the end user. We have our services, as well as several third-party tools and reference designs to enable the load balancing of media traffic across multiple CDNs.
Having two CDN providers as a business continuity and disaster recovery strategy is table stakes for any publisher with a meaningful online presence. Additionally, if the publisher is driving numerous traffic events, such as live sports, promotions, updates and patches, the ability to split the load across multiple CDN providers helps during periods of unexpected surges in demand when capacity is needed to accommodate unanticipated traffic. The primary benefit of a multi-CDN strategy is performance-based load-balancing that takes advantage of the strengths of each provider to deliver a better experience than any individual provider could deliver.
As the number of providers increases, the incremental benefits from increasing the number of suppliers decline as the complexity of troubleshooting, configuration and synchronization increases costs. We see diminishing returns set in quickly as customers increase the number of providers from two to three or four. The cost of complexity more than offsets the benefits of scale.
Another important area for a customer to consider in a multi-CDN environment is how to handle failure scenarios. Disaster recovery should be one of the critical benefits of a multi-CDN solution. However, this becomes complicated when you are talking about moving large volumes of traffic from one CDN to another in a very short timeframe (e.g. seconds). If this process is not well managed, it can cause more problems than it solves -- in the worst-case scenario you can see cascading impacts across the customer origin and the destination CDN and ISP networks.
TT: What advice would you offer online video providers on managing and evaluating the performance of multiple CDNs?
JA: It is important to realize that each CDN has strengths and weaknesses in different geographies, networks and devices and therefore the multi-CDN load-balancing system used should have the granularity to allocate traffic between providers based on the intersection of these three factors. We would recommend that customers start by segmenting their traffic into a reasonable number of quanta, e.g. 20-30, based on the most widely used devices, networks and locations and reviewing traffic allocation between CDNs for each quanta daily. This scheme provides reasonable granularity to compare performance across CDNs without over-complicating the initial analysis.
Once a publisher gains familiarity with the baseline performance across the selected quanta and has a handle on the benefits from switching traffic between CDNs, then they can choose to expand the sets of quanta or to increase the switching frequency. However, our experience is that diminishing returns inevitably set in as the number of quanta or the switching frequency increases.
Standardization of QoE measurement is another area where we feel that the OTT industry would benefit significantly. We find that many of our customers improvise and implement their in-house QoE measurement systems. Unfortunately, each of these systems is slightly different in how it defines and measures QoE. As a result, we as an industry do not have a consistent definition of good experience. Industry groups like the Streaming Video Alliance have attempted to address this gap by defining their standard, but it has not been widely adopted yet.
TT: Level 3 owns both a global CDN and its own IP-network. Are there synergies to be leveraged from this?
JA: Yes, we can combine routing tables, flow and utilization data from the IP network with the application level data from the CDN which includes throughput, latency and errors. These two categories of data add a new dimension to analytics by allowing us to correlate QoE issues identified at the application level to underlying network data, enabling us to find new opportunities for optimization.
Today, we collect data from all network elements and devices and aggregate it within our big data platforms. Our data scientists are now focused on analyzing the data to expand the opportunities for streamlining and optimizing the network.
TT: How quickly are you able to track and analyze video streaming data on your network?
JA: The volume of data is enormous with millions of requests per second being handled on a global basis. We would like to collect and process the data in real time so that we can detect anomalies and adjust as quickly as possible. Currently, the cycle time is running around 15 minutes which is well short of real time. Ideally, we would want to shrink that to seconds.
TT: When you mentioned investments that you are making to prepare for 4K streaming, are you focused more on hardware or software?
JA: We are investing in both hardware and software. The hardware investments are expanding the reach of the network closer to the end user as well as increasing capacity to support the growing bandwidth that higher-resolution video drives. Additionally, we continue to iterate on our server platform to add more compute power and memory for processing higher levels of data at a faster speed. The software investments are focused on supporting increased scale and higher quality streams.
— Kishore Jethanandani, Contributing Writer, Telco Transformation