It is such a central capability to a CDN that it has been said a CDN is, at its essence, “just” a request routing platform.
Traditionally request routing required a deep understanding of the underlying available infrastructure, awareness of where content is, and as services scaled up, the Big Data feedback from end user systems has been contributory to the intelligence on which the route is formed.
In the past few years a number of providers have emerged that were initially focused on performing measurements of the performance of delivery networks to, as we like to say in the sector, “keep them honest” Initially these organizations were termed “measurement companies”
With many publishers - particularly those publishing premium content - it has been common to adopt a multi-CDN strategy. In part, this has been to create price pressure, and in part, it has ensured that they can achieve high availability in the event of a single CDN failing.
The measurement companies started in a position where they created static periodic reports on the performance of the CDNs, which were used by procurement to make weighting decisions about which CDN to predominantly use. Over time the gathering of the Big Data, its analysis, and incorporation into a form that can be fed into request routing intelligence, has become linear, and fast; some would even claim “real-time”
Publishers' value is being able to finely decide how content is going to be delivered from their CDN partners on a per-request basis. However, these same publishers also see this sector as so heavily commoditized that they want to procure the service through a single provider. This in turn has meant that those who have been providing trusted measurement, and even inputting that data to the request routing process have been ideally suited to take over the entire CDN contract from the publisher. So, without running anything more than a request router and their original Big Data systems, they are able to take on massive distribution contracts, with their request router acting as a high-frequency broker, fielding millions of transactions an hour, and to all intents and purposes delivering exactly what the publisher expects from a globally distributed compute network, and picking and choosing which of the actual CDNs gets the traffic and thus generates revenue.
This competition in the CDN space is causing tension by increasing price pressure in a sector where margin is hard to come by. With those brokers also having very low capital outlay to get started (much of the Big Data infrastructure is operated in public cloud) and having minimal operating overheads, they can keep their margin low, and yet play God with the traffic volumes more or less at will - just so long as the end users' experience is good enough. And if it is not, they blame the CDN, tweak their load balancing (between the CDNs), and continue.
Those measurement companies that have taken this particular position are becoming known as CDN brokers, and their market is typically measured in GB or Gbps of traffic - competing exactly with the companies that they were originally “keeping honest.” Being philosophical though, this is also part of the cruel nature of a relatively unregulated economic environment.
The biggest risk the CDN brokers face is that they may make it economically ineffective for CDNs to continue to operate, and if they put a significant number of the major CDNs out of business, then there will be nowhere for the broker to forward a request to, and if that happens, then the model will change.
What is, however, much more likely is that the Telcos will themselves deliver internal CDNs, so the CDN brokers may continue to service nonpremium OTT content, but in the meanwhile the higher value premium content will, most likely, be served on net from infrastructure that will be created virtually and on the fly within the networks for the task in hand. The traditional Pureplay CDN will, most likely, move back from subscriber delivery and be focused on the backbone delivery to the entry point in the subscriber networks' own CDN application - often one that only exists when and where there is a demand in the operator network.