Why are there no mobile CDNs?

If you've been with us so far, you probably know that Telco 2.0 likes content-delivery networking (CDNing) a lot. So much so we invited an executive vice president from market leaders Akamai Networks to the next Telco 2.0 executive brainstorm this October. Even though a couple of CDN operators - Akamai and Limelight - recently had a bad day on the stock market, we're still confident of this judgement. After all, when Wall Street is annoyed because your profits were only up 55 per cent, it's probably their problem rather than yours.

A CDN, to recap, is a way of delivering bandwidth-heavy content (usually video) over the Internet efficiently. Standard methods have the downside that the same material has to be transferred out of the provider's network, over the backbone, and into the user's ISP network at least once for each user; peering and transit costs are incurred at each stage. Further, the load on the provider's servers is a problem. In a CDN, Web servers are placed at strategic points inside customer ISPs and filled with content. Requests are then redirected to the CDN box, so each item only has to be transferred outside the ISP once.

It has the major advantage that everyone's happy; you're happy because the load on your own servers is relieved, and your stuff is delivered faster than the competition, your ISP is happy because their peers aren't yelling any more, and the downstream ISP is happy because the weight of traffic has been moved inside their own system, where (depending on their business model) it's either cheaper or effectively free. And none of this involves reducing other people's quality of service or doing any other evil. It's elegant engineering, and good economics.

The reason, deeper down, why CDNing works is that it understands where the bottlenecks are; the critical paths, the limiting factors, the maximum cuts. In this case, the bottleneck is the network edge, because it's where the economic activity occurs. The Internet's economic model is that interconnection is what creates value, realised either by barter (peering) or in cash (transit).

So why are there no CDNs in mobile networking?

Traditionally, it was because the bottleneck is in the wrong place. While the mobile radio link was really, really slow (and dodgy), there was never going to be enough traffic to necessitate CDN, and anyway, there would be little benefit in it as the CDN would be on the wrong side of the delay.

But now? In our 40Gbits Granny post, we mentioned the scale of capacity upgrades some mobile operators are putting in; upgrades that were triggered by introduction of HSPA in the radio network. With credible (megabit-class and above) data rates, the operators' grandiose plans look..less grandiose, and the challenges rather different.

Now, mobile operators tend to be very keen on rich media content; video and music. With the bottleneck in the radio network removed, it's time to think about how to deliver this stuff. The first problem is backhaul; the same operators buying Cisco CRS-1s are also busy laying fibre to their cell-sites. This raises an interesting question; what about CDNing below the level of the carrier? Why not bring the content further down the stack, closer still to the users? CDNing is rather less appropriate for operators who are paying per-byte for backhaul, unless the CDN server is right at the base station, but it can still save on backbone and internal transmission costs. For those operators who own substantial backhaul assets, though, it's near ideal.

The second problem is core-network latency; one of the causes of which is that the traffic has to go anywhere near the core network in the first place. In ISP-land, the practice of having multiple cached DNS (Domain Name System) servers near users is so common as not to excite comment; but quite a few mobile operators force all queries through their Mobile Switching Centre. As each webpage may require multiple DNS requests, this can cause more delay than the radio link.

So, in order to speed things up, mobile network vendors are now offering products that break-out traffic to the Internet closer to users, as low down as the Radio Network Controller (RNC). If you're bringing IP that close to the radio network, it begins to sound plausible that other things might join it; DNS caching, and then why not heavily-demanded content? Further, a lot of mobile networks already push rich content through "content handling" servers of various kinds to squeeze it down into more compressed formats. So why not CDN as well? Where there's a bottleneck, there's usually an opportunity.

Telco 2.0 is about moving to a horizontal, modular business model. Whilst much of this is driven by the Internet and pressure to become a "dumb pipe", the reality is that we are moving to a rich and complex network architecture. Traffic shaping, storage and processing power are embedded at many points between central data servers and backbones to edge radio networks and devices. A successful operator will be the one with the right mix.