Preview: TYR @ OpenTech 2017

One thing I’ve recently noticed knocking about the industry is what I’m beginning to call the “post-Internet” or maybe the “Afternet”. This is the conjunction of maybe three trends.

The first of these is the way 5G is panning out over in mobile.

To start with, pretty much everything in consumer and most things in enterprise are mobile these days so there’s no excuse for ignoring those boring 3GPP suits. Also, it looks very much as if the first deployments of the 5G New Radio are going to be fixed-wireless, as a replacement for DSL and a substitute for point-to-point microwave or maybe some fibre. So rather than being a “mobile thing”, they’re going to be another access network medium for the general public Internet service.

I’m not going to go into the radio stuff here, fascinating though it is. What interests me is that the 5G builders are increasingly keen on integrating general purpose computing capacity into the network, and importantly, putting it within the radio-access network or at least the mobile core network. So you might host an app inside T-Mobile or Orange Armenia or whoever’s RAN, which would obviously be lovely from a webperf point of view as a way of getting latency down. And, you know, distributed systems design is fun! Latency is a huge deal in 5G; 3GPP wants a round-trip time from the User Equipment (yes, they still call it that) to the carrier’s edge router facing the Internet of 10ms. The important point to bear in mind here is that if this is delivered, the competitive imperative to get the full benefit of it will be powerful. We’ve known for years that in all kinds of applications, latency and its standard deviation have a direct impact on user experience metrics and indeed on business KPIs.

This brings us to the next trend. I recall not so long ago Jamie Zawinski saying on his blog that Instagram was barely on the Internet at all, in any meaningful sense. At the time they didn’t have much of a Web site, so it was almost entirely an app store/smartphone experience. You couldn’t just download or upload stuff from whatever computer you happened to have with you. And you know, if they could host at least an upload accelerator reverse-proxy inside the mobile networks they certainly would. CDN nodes for the pix would also be valuable. Akamai will kinda-sorta do this for you but the actual coverage is pretty thin compared to what it is for fixed eyeball ISPs. Seeing as Google already has an absolutely huge global caching/CDN infrastructure, I’m surprised it’s not an Android API already.

Now for our third trend. If there’s one product that goes out supremely well in the enterprise these days, it’s Layer 2 Ethernet service, as opposed to IP-VPN or whatever T/E carrier grandad’s still using. One of the major applications for this is that some of the providers now have L2 interconnection (called an E-NNI for External Network to Network Interface, if you want people to think you’re a Bellhead for some strange reason) with the big clouds. So you sign up and suddenly MS Office 365 is sooo much snappier. From their point of view, this is really great if you’re a huge US telco or cableco that can get all its enterprise customers and three AWS Regions on-network, but not so good for everyone else. So the smaller operators – and some not-so-small but multinational ones – are trying to sign up as many E-NNI agreements as they can with each other, standardise their internal IT, all in the aim of being able to increase their L2 footprints.

The dark side of this is that in doing so they’re shifting emphasis from the Internet’s Layer 3, multilateral peering interconnection model to a new L2 private peering ecosystem. Personally I’m not sure it can work – one of the underestimated features of the Internet is that its federal nature means it has clear administrative demarcs, and this certainly won’t, as it’s going to be some sort of turtles-all-the-way-down but with VXLANs protocol tunnelling nightmare, which will make debugging it pure hell. A preview of this is the way a lot of the big clouds really struggle to do networking at all well.

The opposition, meanwhile, seems to be companies like Google who run proper IPv6 networks, but increasingly own everything but the last mile and increasingly rely on their own in-house fibre for everything. Google specifically has every reason to obsess about this after that time the NSA tapped all their inter-datacentre WANs and they bought the world’s supply of FASTLANE linerate encryption boxes all in a week.

Hence the Afternet – an oligopolistic collection of semi-autonomous clouds overlaying politicians’ demands for censorship and digital protectionism. Will your blog be on it? So, where are we going with this and who should we subject to remorseless public shaming? Well. That’s why I’m going to be speaking at this year’s OpenTech on the subject of 5G and Your Website. You may have noticed there’s a link to “5G Resources” at the top of the blog – there is nothing there at the moment but I promise it will be full of links before the 13th of May.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.