Hyperbola, Utopia or Just a Couchemare?
Wherever you look you will find predictions about ‘edge’ data centres, from the serious to the downright silly. However, the concept of edge computing is simple; the data traffic coming from the ‘Internet of Everything’ will overwhelm the communication networks and putting the compute (and short-term storage) closer to the load will take the local strain, cut down the latency (delay in communication) and be able to filter out the valuable data to be transmitted onwards and upwards to huge hyperscale data centres which will carry out all the ‘big data’ things – including (if you are cynical) selling our preferences, prejudices, moral weaknesses, movements and spending habits etc to anyone that wants to know about how they can dislodge cash from our pockets. There will probably be a middle layer operated by our present collocation data centres where the connection from all the small, numerous and local ‘edge’ access points can be consolidated into a converged network of fat-fibre.
The Internet-of-Everything means exactly that; smart-buildings, -transport, -grids, personal video entertainment, numerous ‘phone’ apps accessing the internet and smart-wear like watches and glasses etc. Too many to mention, half of which we can’t predict yet. The perfect edge application is the driverless car, clearly with lots of on-board computation capacity, but also needing a wider vision of its world – not least traffic data for routing and road hazards to avoid, all with split-second. Would you feel comfortable as a passenger knowing that your safety depended upon a data connection to another country or inside the arctic-circle?
Another application has already started to follow the road to edge; YouTube dominates the internet traffic of the world and viral-videos could choke the network. But no-one wants buffering for longer than a second (or nanosecond for some) so Google already push the most popular content to rented collocation facilities in the metropolitan areas to limit the long-haul from their remote behemoths. But its simpler than that – we are not far from streaming UHD movies on our phones whilst travelling on the Underground, how could that be possible without a local source doing the buffering in the background for you with lovely advertising breaks?
So, how many ‘edges’ could there be? The answer probably lies in a technology roll-out we have already experienced: For mobile phones why don’t we have one huge aerial in the middle of the country (Banbury?) to which all 80 million SIM cards connect? Or for fixed-line phones why not one huge telephone exchange? No, we have more than 40,000 mobile-phone masts, each serving a small area (remember they were called cells?) and densest where the users are. Edge will be the same – densest deployments where the greatest number of applications (us) are. In a city like London, probably every 500-750m?
How ‘big’ should they be? The ‘smart’ answer could be ‘40% smaller year-on-year if the load doesn’t grow’. In other words, the effects of Moore’s Law (in all its forms and derivatives) shrink everything over time. The only reason that data centres have grown is that demand has continually overhauled capacity – consider the 2MW (2,000 x 1kW cabinets) ‘internet data centre’ built in the year 2000 that can now be outperformed by one 5kW cabinet. That is Moore’s Law in action. So, anyone who predicts that edge facilities will be ‘X’ kW or ‘Y’ cabinets or ‘Z’ servers is very likely going to be wrong. I don’t mind being proved wrong, so I would guess that 20kW of the best hardware should suffice at the beginning and then refresh the components every 18 months. When I say ‘best’ I really mean fastest since the difference between the best and the worst is a factor of 150:1 (read my last article in this journal) so that ‘X’ kW can mean everything, or nothing.
The other point about ‘size’, or kW capacity etc, is that the edge facilities will serve very finite areas with a certain population density, so that demand goes up it could be served by another edge installation, not just a fatter ‘edge’.
So why 20kW? Well, we come to what I believe will be a key feature of ‘edge’. We are all concerned about the energy that data centres consume and so the thought of having 80,000 20kW facilities added to the power network (1.6GW at a PUE of 1.2 = 2GW = 5% of the UK grid) is not without concerns. So, the obvious advantage is to reuse the waste heat and make the edge facilities grid-neutral. All we need is a nearby 24/7/365 heat load… Clearly the solution is to liquid-cool the 20kW of ICT load and take off the heat as hot-water at 65-70°C. House the ‘edge’ in a cheap hotel room (which already will have a 5G aerial on the roof and a fibre connection), hospital, care-home, leisure centre or even apartment block and they can absorb 20kW (and a lot more) of hot-water for domestic services, 24/7. The host gets the rent and the hot water whilst the edge operator gets the ICT revenue, pays for the electricity and maybe even charges for the hot-water. The internet becomes carbon neutral, even ‘green’ for the first time. It is little different to city-centre buildings that already rent their roofs to the mobile phone operators.
Just like mobile phone cells, if one goes down the users experience almost no difference as the nearest in the cellular network cover the area, albeit with a weaker signal. But ‘edge’ will be smaller cells and so redundancy in service coverage will be more that sufficient. With ICT loads already capable of remote diagnostics, upgrades and re-boots (as most collocation users already can) the edge facilities will dark and unmanned. We will create 60,000 new jobs (20,000 on three-shifts) of mobile technicians. With the skill shortage we already have these mobile techies can start being semi-skilled – wearing cameras with a headset and 24/7 access to a technical help centre (hopefully not overseas) where fully skilled technicians in each subject area, power, cooling, ICT etc can guide the on-site technician ‘live’ on camera and head-up display. It does sound rather 1984, doesn’t it? But to us oldies so does the idea of a phone that is used for everything other than making phone calls – does anyone else remember how advanced we though the ‘bleeper’ was? And then came along the bleeper with a single-line 16-character message display…
And just when you think you have a plan, along come the silly ideas… Robots serving 9’ high cabinets? Well, 42U will be more than plenty for 20kW of compute and storage plus a network switch and, anyway, 9’ high cabinets won’t fit into the average Travelodge or Holiday Inn Express (other hotels are available). I assume the robot is there for the height or pushing reset buttons? All they will provide is 60,000 more jobs for fixing them when the go down. No, the best solution is to rent the nest door bedroom in the hotel. That way you can isolate the annoying fan noise from other guests and use the room to let the on-call technicians to sleep in when on night shift – less CapEx than a robot and much easier to train…
Finally, we could not finish any discussion about anything ICT related, especially ‘edge’ facilities, without mentioning ‘AI’. Again, simple. We are 50+ years away from AI (as defined by Alan Turing) as everything we now call AI is built on algorithms written by people and compiled on machines that use software written by people. The junior brother of AI, Machine Learning, is also based on code written by people… People are not artificial, yet, but the key word is ‘people’ and, at the end of the day, that it is one resource that we have too much of and the last thing the economy of any country wants is unemployed people.
Make no mistake, ‘edge’ is logical, sensible, required and unavoidable and gives us a great opportunity to get the re-use of waste ICT heat, at last, possible for cities that don’t have district heating. That is, most cities…