PNSol: Answer to the broadband riddle?
For some unknown reason, I have this mental image of 1940s actress Hedy Lamarr, glamorous co-inventor of spread spectrum radio, doing her work in some swanky Manhattan hotel suite. A grand piano sits in the corner, and servants flutter by. Who wouldn't want to be a fly on the wall as an eye-catching actress chalks up one of the foundation stones of modern communications? And all over afternoon tea and cake.
Wait a few decades for the invention of the transistor, mix in some incomprehensible algebraic magic, et voila - CDMA radio, 3G and footie clips on your mobile.
In stark contrast, we've been down in an anonymous terraced house in Clapham in south London, quietly watching agape at a technology that could unleash a revolution of equal magnitude, only this time targeted firstly at fixed networks. Just as with CDMA, a bunch of boffins have applied some clever maths, and worked out how to get a ton more value out of your communications network.
The company is a tiny start-up, Predictable Network Solutions (PNSol). And it's got "disruptive" written all over it in big neon capitals. We don't have any shares. They've not paid us anything (although they're generous with cups of tea). We're just fans.
Broadband: nice for users, nasty for investors
Before we look at what they do, let's take a moment to review the seemingly intractable problems of the broadband Internet market, since no greater love hath capital than to lay down its life for a competitive broadband access network.
In the days of vertically integrated products like voice and SMS, traffic, cost and revenue were tightly coupled. If you used a hundred times as much of the network's capacity as your neighbour, expect to get a rather larger bill. Then we moved to dial-up Internet, and all of us became email and Web addicts (at least between 7pm and 11pm, until our bosses later understood how critical reading Dilbert at work was to our morale and thus productivity). We used just a few text-centric applications, and all had similar usage profiles. Flat rate pricing kept us with a need for speed, craving ever more capacity.
Then some people started snorting up gigabytes of P2P traffic, and others mainlined streaming video into the night. A small proportion of users had an online habit that had spiralled out of control. We'd all love to live in a world of end-to-end fibre, user-controlled lightpaths, and infinite capacity. The reality is significantly capacity-constrained copper almost everywhere. Your greedy behaviour drives a poorer experience for other users (just ask any TalkTalk broadband user if they're enjoying the advertised 8 Mbit/sec in the evening...), and capex and opex by the ISP.
ISP = Internet Solvency Problem
The ISP's problems are thus:
- Cost inflation due to P2P and streaming video traffic, but inability to raise prices enough to compensate.
- Divergent and unpredictable behaviour of users, causing inability to rationally price the product, segment users or shed the hogs.
- Revenue stagnation due to inability to tier traffic or guarantee quality of experience for voice and video. Skype works, but it doesn't consistently work; many high-value business applications don't fully migrate to broadband.
- Technical and political push-back from existing traffic shaping approaches. There's a cat-and-mouse game between ISPs and their users.
- Hideous cost and lack of revenue model for IMS-like technologies that try to put the open Internet genie back into the bottle.
PNSol solves all of these by simply extracting the maximum possible value from a broadband connection, rather than over-provisioning the network (often implying massive capital expenditure). It doesn't miraculously create new capacity, but lets you squeeze every bit of value out of what you've got.
The user also has a problem. When they are doing a VoIP or video call, and their kid fires up BitTorrent or the BBC iPlayer, this will necessarily contend for capacity and degrade the call quality. At present this is tricky to manage. Even if you prioritise one flow over the other in the home router, you are a long way from exploiting the total network capacity because these packets will remain contending with one another in the ISP's networks. PNSol can solve this too. Your BitTorrent traffic has zero effect on the VoIP traffic. Indeed, it doesn't even need to contend for capacity at all. No capex upgrades just to keep the Skype users happy and their heads above the sea of P2P.
A selectively stupid network
So where to start about PNSol? Well, a good place might be to take a look at our post from early last year on Paris Metro Pricing. What this does is divide up the broadband superhighway into "lanes", and then charges different amounts for each lane, and lets the applications firing the packets dynamically decide which one balances price, delay and reliability.
PNSol strongly fits with our "slice and dice" and rich wholesale vision of how the broadband market will develop. Specifically, it allows for the "dicing" by quality, not just the "slicing" by quantity -- and does it in a way acceptable to both users and ISPs.
Essentially, it's Paris Metro Pricing with a twist: not all carriages are the same.
How does it work?
Their technology comprises two parts:
- Software to go into the home hub, which pre-conditions traffic before it enters the ISP's network.
- A network element to go in the core of the ISP's network that manages inbound traffic to the subscriber.
Unlike today's traffic management approaches using DPI, PNSol does not allocate "bandwidth" or manage individual user sessions. Instead, it uses an advanced statistical model (repeat after me: stochastic finite-state automata) to allocate a budget for delay and loss between packet streams. Unlike with DPI, there's no need to open up the packets, and it doesn't matter if they are encrypted. The user's can't cheat by mis-representing their packets (e.g. tagging 'premium VoIP' as cheap P2P), since those VoIP packets will get the delay characteristics of the P2P ones (i.e. unusably slow and jittery).
What's the secret sauce?
They model the properties of the broadband network itself, and carefully control the packets that are allowed onto it. They ensure that no queues build up within the network, and there are minimal losses within the network. The system works because although you don't control the telco's network (e.g. BT's IPStream product), it turns out to be highly stable and predictable. Thus they can guarantee the application that it will get exactly the throughput, delay and loss characteristics is needs. It sounds simple, but it's not.
Say that again, in English
If you've driven on the M25 orbital motorway around London, you might have come across the variable speed limits that smooth out traffic flows during peak periods. A step further is the entry control system for the M42 around Birmingham, which has traffic lights on the slip roads. Well, imagine that your car was picked up by a crane and inserted into the slow, middle or fast lane at just the right moment and at the right speed, such that congestion was minimised and road usage maximised. And then plucks the cars off at just the right moment too. That's kind of how it works.
Does it work?
Yes, PNSol have set up their own ISP on BT's wholesale network, with the kit at the Telehouse exchange in London. We saw superb 2-way video over a standard 2mbit/s DSL line, including to someone else in the UK on a 512kbit line (with full duplex video), and also to someone in the US. Far better quality than you'd usually see. And you can flick the BBC iPlayer on and off ... with no effect on the real-time traffic.
So how does it solve the ISP's problem, again?
- The user can run P2P as a 'scavenger' class of application, with huge total bandwidth and extremely low cost. This is because the traffic is pre-conditioned in a way that never increases the peak load on the network, and thus never causes any capital expense. You've only got the opex of the electricity to run the routers...
- The ISP can offer premium tiers of connectivity (by quality) at premium prices. A 2-way video stream is expected to cost 1-2p/minute based on current DSL wholesale pricing economics. This does not require the end points to speak any special IMS dialect. Indeed, they don't need to talk SIP. Just need to make themselves identifiable using a port range or quality of service flag in the IP header. Thus the range of compatible applications is very high.
- There are few network neutrality issues, because the user is ultimately in total control. You're just re-allocating the available capacity into multiple virtual networks with different delay, loss and throughput characteristics. The user decides what traffic goes at what priority. There's no telco as 'Internet highwayman' because it doesn't matter where the traffic is ultimately headed, or even what it really is. And the user has to live with the reality of jitter and latency anyway, so we're giving them more than what vanilla best effort offers by offering them control over how that degradation is allocated.
Not all revolutions are welcome
PNSol technology won't be welcome everywhere. If you're selling big routers, your salesmen don't get to go to Hawaii and the Chairman's Club this year. Or next year. No need to throw capacity at every network contention problem. You can get a ton more value out of today's kit. And an awful lot of NGN products look obsolete before launch. Why buy super-expensive IMS kit to reserve (99% empty) capacity on the network then "better than best effort" is available at a fraction of the price? It also doesn't look good for MPLS-based networks, and could force a deep re-think at a lot of telcos, especially around the small business market.
The ISP re-education camp beckons
It's going to be tough for ISPs to transition to this new kind of technology. How to package it up and retail it? [Ed - we're here to help...] It's a long education process as it involves thinking about the network in a new way. There's no such thing as "bandwidth" - it's just a colloquial metaphor. You're in the business of maximising value of a network by pricing loss, jitter and latency among the competing uses and users: a logistics company for data, matching bulk, urgency and value to the available transport.