The traditional CDN technology is great, but its main drawback is that it requires up front investment to cover an unknown traffic peak level. Popular live events like the FIFA World Cup or season premiers like Game of Thrones are recurrently reaching new traffic peaks exposing failing CDN performance.
In this third and concluding article about OTT content delivery, we explore various types of CDNs, alternatives to CDNs, and aspects when building or buying CDNs.
This is the third CDN article concluding the content delivery topic. The previous articles describe what CDNs are, how they work, and how multi-CDNs are built.
This publication is a part of a series of articles describing the principles of the technology behind video streaming. It could be read without any prior knowledge on the subject.
Types of CDNs
There are generally four types of CDNs, all described below. Three of them are built on technologies and principles already described in the previous OTT Content Delivery articles. Brief classification is described below. The fourth technology for content distribution is built on different principles and is therefore described in a separate section below.
Public CDNs offered as a service for content distribution can be divided into two subgroups: Enter-Deep architecture and Super POP architecture CDNs.
Enter-Deep Architecture CDN
The Enter-Deep architecture is based on distributing the CDN pops as close to the clients as possible. The POPs are usually placed further out in the access networks enabling a very low round-trip-time to clients. The performance is usually very good and the resiliency against D-DOS attacks is great with many POPs. Business wise this approach requires a good cooperation between the CDN and ISPs owning the access networks and allowing CDN POP hardware being placed in ISP premises.
The advantages with lower latency and performance is usually counter-balanced when maintaining the high number of POPs. Configuration and changes propagation may be challenging. Also, due to the many number of POPs, it gets too expensive for each POP to hold a large cache or even all functionality. Smaller caches create a higher number of cache misses and thus generates downloads from other POPs in the CDN network.
Super-POP Architecture CDN
The super-POP architecture aims to intelligently place fewer but much larger POPs in strategic locations like exchange points or large peering points. These high performant POPs are very well connected to high bandwidth links to multiple ISPs and can therefore achieve a good performance towards clients.
With few POPs the maintenance and configuration propagations are rather effective. Also, the larger POPs make sure that cache misses are kept low. However, fewer POPs require carefully planned POP locations for optimal connection performance and peering agreements efficiently reaching any client in the region. And if any of the few POPs fail, the clients will be directed to other POPs usually much further away from the failing one, potentially generating higher latency.
Proprietary CDNs are built exclusively by and for network owners like operators, ISPs etc, and handles own traffic only. The POPs usually reside within the owner’s networks or rented POP locations. By building own CDNs the owner is free to choose any technologies and tools to address any network prerequisites or business challenges. The proprietary CDN can be perfectly dimensioned and built for the target group. Total CDN control, full access to all statistics and usage data, and no risk of CDN congestion by other users, are some of the advantages of proprietary CDNs.
Building private CDNs requires lots of technical knowledge and continuous work to make sure that any traffic fluctuations are prepared for. And traffic peaks may become a scalability nightmare since they come rarely but nevertheless require up-front investment and preparation.
Being very CAPEX heavy, proprietary CDNs usually become very cost effective operationally with a steady base traffic load.
The media streaming industry often relies on multiple CDNs (see last article) or even Hybrid CDNs. A Hybrid CDN is simply a mix between public CDNs and a self-built CDN, and enables a perfect combination of control, cost, and performance.
Hybrid CDNs may be used for various improvements to content distribution. Improvements like enhancing performance in certain regions by adding own POP locations, increasing control over certain streaming services by for instance pushing VOD traffic to own POPs, or simply add a supplementary remote storage location. Any quantity of self-built components shapes a Hybrid CDN.
There are few alternatives to CDNs that can offer a matching scalable delivery solution to millions of subscribers, various geographies, or the availability requirements. Distributed caching points in the network seems to be the only solution so far. However, there is another distributed cache node infrastructure available to anyone — the client population. Mobile devices, TVs, Set-top boxes, and PCs are all downloading and caching content to consume it.
P2P (Peer-to-Peer) CDNs
The P2P (peer-to-peer) technology, for some mostly associated with illegal file sharing, can be used by content distributors and video service owners to offload CDNs by having clients sharing the content. Unlike the traditional server-client model, a P2P network consists of equally privileged peers (video clients like mobile devices, TVs and PCs in our space), all being both consumers and suppliers of resources. Each of the peers in the network agrees to share part of its hardware and connection resources to cache and share content to other peers in the P2P network. More details about the P2P technology falls outside the scope of this article and may be found anywhere on the internet.
In P2P content distribution origin may be tremendously offloaded, vendors in the market talk about figures up to 90%. Instead of downloading video segments from a CDN, why not just download them from the other client? The more popular a live event is, the more clients are involved in the content sharing. And the traditional CDNs remain unaffected by the traffic increase.
As previously mentioned, P2P technologies are used to offload traditional CDNs rather than being the single solution to scalable content distribution. The reasons are simple; exposing content origin servers allowing consumer clients on the open internet to directly request content files would be a serious security threat with probable DDOS impact.
What’s Holding Up the Parade?
Why don’t we see P2P networks everywhere?
There are a few important aspects holding back the quick adoption of P2P content distribution. Traditionally the issue has been that P2P CDNs require subscribers to give up part of their device resources — upstream traffic, local storage, CPU cycles and battery consumption.
However, some P2P vendors are able to configure their technology to use only clients on fixed connections for content sharing. This usually solves most of the concerns since fixed clients are usually more well equipped, especially on power and connection. It also solves a much more alarming problem; mobile networks are usually not dimensioned to have upstream traffic added. This could potentially threaten the downstream traffic which is unacceptable. But again, having only fixed connected clients sharing content, there is no need to worry.
There are more worrying disadvantages with P2P content distribution — disadvantages that are becoming more apparent as new video related services are emerging. Following two major services are today not compatible with P2P content distribution:
Server-Side Ad Insertion (SSAI)
Server-Side Ad Insertion, described in the Ad Insertion articles, offers a way to replace ad commercials in video streams and VOD content. As the client is completely unaware of ads with Server-Side Ad Insertion, the benefit of inserting a consumer specific targeted ad gets lost, since the content is downloaded from other clients. All clients in P2P content distribution will receive the same ads.
Users of P2P content distribution are left to use Client-Side Ad Insertion, in which the client is ad aware and explicitly asks an ad server mid-stream for ad content, before it resumes playing the video stream. This is however one of the main disadvantages with Client-Side Ad Insertion. As the client is aware of the ad break, it is susceptible to ad blocker software.
Read more about Ad Insertion technologies in the Ad Insertion articles. It is interesting.
Content Watermarking, described in the Securing OTT Content — Watermarking article, is still not compatible with P2P content distribution. Content watermarking, used to uniquely mark each stream to identify the original owner, gets disarmed since video segments from any number of clients are blended to a consumer stream. The original owner, potentially sharing live streams illegally, can no longer be identified.
Priority 1: Service Availability
At the end of the day, P2P content distribution may be the best technology today that can prevent the recurring problems with CDN content distribution during the highest traffic peaks. Any major sports event or premier teaches us the same lesson: that CDNs may break leaving the service temporarily unavailableThe technology is undoubtedly interesting but has important unresolved challenges to address.
The Joker: Multicast Distribution
Looking beyond distribution infrastructures, there is currently an emerging technology that may prove to be very efficient for streaming live events at massive scale: Multicast.
Multicast is not a new invention. Multicast, or streaming one-to-many, is an old well proven technology used by many applications. In the video space, IPTV (described in the Streaming IPTV article) has always used multicast to stream live content efficiently to any sized audience. However, network protocols associated with multicast streaming has so far been suited only for private or closed networks, and thus unsupported on the open internet. Besides, adaptive bitrate streaming, explained in the Internet Video Streaming — ABR articles, where clients adapt to network fluctuations has so far been incompatible with the multicast principle.
Current research projects are breaking the barriers between the two incompatible technologies enabling adaptive bitrate streaming over multicast. Streaming major live events over multicast will reduce server resource consumption (for live streaming) to close to zero.
Another initiative; LTE Broadcast (a.k.a. LTE Multicast) enabled by 3GPPs evolved Multimedia Broadcast Multicast Service (eMBMS), already makes Multicast content streaming available within eMBMS LTE mobile networks. While the technology emerged already back in 2014, it has still not taken off properly. The technology must be explicitly implemented in mobile devices and mobile networks. Samsung was an early adopter on the client side, but only a few global operators have so far added the support. Critics point at slow network upgrade cycles and that the rollout cost can’t justify the business case. Besides, many are probably waiting for the 5G network rollout
Either way, Multicast for ABR streaming is ground-breaking enough to be covered in a separate article.
Aspects of Build vs. Buy
Content distributors today need to decide whether to buy a CDN service or to build and maintain an own CDN. Building a CDN is high on CAPEX and low on OPEX (of course depending on the organisation). Buying a CDN service is the opposite — low on CAPEX and high on OPEX.
Let us have a closer look at the business cases of these two alternatives.
Buying a content delivery service from a CDN provider is usually straightforward. The price model is simple: usually per GB content delivered, and potentially some fixed starting costs for the account and certificate setup. Minor configuration of DNS is needed.
Setup and configuration can be made within hours or days and no technical CDN knowledge is required.
Building an own CDN is much harder and requires a deep network knowledge. The CDN network needs to be dimensioned, server hardware purchased and installed, and structured as a CDN network. Peering agreements need to be dimensioned and established making sure that the service can be made available to clients connected to any ISP.
A privately built CDN must be continuously maintained. This is a common misconception; a complete working CDN will not run by itself forever. It has to be maintained. The traffic flow varies constantly, as the underlying network infrastructure (routers and connections) and other parameters.
Last but not least, don’t forget the hardware depreciation. But hold on, this may actually be a possibility. Building CDNs with public or private cloud technologies, owners may scale their CDN on demand and use available system resources for other tasks during low traffic periods. The platform is still required for other IT related services offered by ISPs, Telcos and others. Adding the CDN application as another component on top of an already existing cloud infrastructure may be a wise option.
Fact is that the CDN prices are dropping fast and the multi-CDN technologies described in the previous article OTT Content Delivery — Multi CDN are accelerating the plunge. Read Dan Rayburns excellent blog post about this:
The cost saving in building and maintaining an own CDN is rapidly shrinking. In parallel, decreased revenue per GB delivered is made up by including other technologies with the CDN. The public CDN market is going through heavy consolidation. Technologies such as more sophisticated security offerings, AI applications specifically for video (and other areas), and other video related services like storage, transcoding and packaging etc. are bundled with CDN in a more complete offering.
There are several compelling technologies for video content distribution available to both new market entrants and veterans. Choosing a perfect combination of the technologies, and keeping an eye open for multicast ABR distribution, will surely keep any video service highly available and with top quality.
Building an own CDN may look intimidating and expensive, but with a stable base traffic flow a private CDN may still be cheaper and offer better control than buying CDN services. Don’t forget, the public CDN companies are in fact building their own CDNs. The only advantages they have is their CDN expertise and probably a generous hardware volume discount.
To minimize the risk when starting a new video streaming service, keep the CAPEX low not to end up with lots of infrastructure. And in the consolidating market, make sure to keep all your streaming platform options in perspective rather than considering CDNs and other components separately.