67% … 84% … 93% … 99.2% … and then stuck interminably at 99.9%. C’mon! Download you unspeakable nuisance! I need to see the next episode.
We might think we still see the dread message “Buffering…” far too frequently, but in reality in any major metropolitan area we are already spoilt by extremely good 4G coverage. Streaming high definition content from video on demand rarely requires us to endure a spinner, progress bar or percentage counter for more than a second or two. This kind of access to near-ubiquitous high bandwidth network connections would have seemed miraculous half a generation ago.
There are three numbers that are particularly important when deciding how good a connection is: download bandwidth, upload bandwidth and latency.
Download bandwidth is easily the most familiar of these three – we’re used to Internet Service Providers touting the download bandwidth of their services: “16 megabits per second ADSL”; “50Mb/s Fibre”; “300 meg Ultrafast Fibre”. The higher the better. Often upload bandwidth is a fraction of the download number – 12 megabits per second up alongside 100 megabits down would not be an uncommon ratio. The logic is that most users will tend to download much more than they upload, and therefore biasing the total available bandwidth within the user’s connection toward download speed will generally best match their use.
Latency is a measure of the responsiveness of the connection. If the user uploads a simple request (a simple “ping” request asking the server on the other end to confirm that receipt of the request for example), the latency is the shortest period within which they can expect a response. Response times are measured in milliseconds, so a ping response of 43ms means that the fastest response of that connection is forty-three thousandths of a second. In general, humans tend to notice network delays when they reach around about 100ms, or 0.1 seconds, and today’s 4G networks have an average latency of around 53 milliseconds.
Simple consumption of content from the network – streaming video being a prime example – is generally more sensitive to bandwidth than latency. Latency might introduce a slight delay in the video starting, but having a high download speed will ensure that the stream continues.
But for many kinds of use the responsiveness of the system is key. High latency would be very noticeable.
One field where responsiveness is paramount is gaming – the split second “twitch” responses required by the best gamers mean that milliseconds’ difference in latency makes the difference between a winning and getting ‘fragged’. The most dedicated PC gamers have long spent ever increasing sums on building and upgrading their computers with faster CPUs, massively parallel graphics capabilities, the latest generations of RAM and storage in pursuit of the most cinematic graphics and highest frame rates. All of this requires an investment of thousands in hardware upgrades, and tinkering with overclocking settings in software, and even so the system then sits idle and unused for a significant portion of its life.
When playing games online, all the game code runs locally, and player’s machines exchange small amounts of data over the internet (about other player’s position, movement, shots etc.). Latency is still an issue, and gamers want a low-latency internet connection, but the fact that most of the work of running the game is done locally (and predictive technologies within modern multiplayer games that “fill in” where other players are likely to be) means that networked PC and console gaming is actually reasonably tolerant of latency issues.
Google’s recently announced “Stadia” gaming service turns the traditional gaming model on its head. Instead of having a high-power machine locally, low power devices simply connect to Google’s servers and display the output to your TV. They receive streamed output from the a virtual machine running on the server, as if receiving streaming video from YouTube. The gamer plays the game using a gamepad that connects over the Internet to the same virtual machine. Instead of paying thousands to own a PC, and having to spend hundreds more each year upgrading it, gamers are offered the attractive prospect of buying a cheap internet appliance and a monthly subscription to the platform – gaming as a service.
Just like the old ‘client-server’ computing models of the mainframe era, the user has what amounts to a ‘dumb terminal’, and all the computational heavy lifting is done remotely. This has the advantages that machines can be better utilised as part of a global service (if gamers mostly play in the evenings, the machines can be near-fully utilised with Asian gamers gradually replaced by European games followed by American gamers in each 24 hour cycle), and with the machines close to each other in a data centre, latency between players in the game will be incredibly low. However, latency between the player’s screen and the remote computer is critical.
Stadia isn’t the first service of this kind – others have tried game steaming, but none have succeeded. Perhaps the highest-profile previous effort was OnLive, but that service shuttered in 2015. OnLive and services like it never managed to solve the latency problem. Games complained that using these services there was too much lag – too long a delay between an input on their controller and the reaction on screen.
Google is in a very different position to previous companies attempting this type of service. Unlike many other start-ups who have tried to crack this problem, Google already has enormous experience in creating highly responsive massively parallel internet services, and are used to serving complex content to users who demand responsiveness. Everything from searches, to YouTube to its Google Docs platform.
Perhaps more importantly, it is launching this service just as 5G mobile services are launching. Whilst much of the publicity about 5G has focussed on its increased download and upload bandwidth, another feature of the service is its much lower latency when compared with 4G and earlier mobile technologies. We know that current 4G bandwidth can easily sustain a 4K ultra-high definition video stream, and low-cost TV dongles (from Roku, Google etc.) exist to play such streams. If 5G brings lower latency connections (within the realms of the typical signal-to-screen latency of many computer displays) then terminal-server gaming will easily satisfy the needs of all but the most elite esports gamers with their apparently superhuman reaction times.
A swing back from local computing to terminal-server models could also revolutionise many other computing environments. Instead of buying a home PC for office tasks, internet browsing, image or video editing and other typical home computing tasks, a Stadia-like system (but with a keyboard and mouse instead of gamepad) can just as easily provide a platform for those tasks too. In many ways that is the ultimate fulfilment of the Chromebook and other ‘cloudtop’ machines.
And what’s good for the home is just as good for enterprise. Ubiquitous ultra-low latency and high (enough) bandwidth connections will be attractive replacements for many businesses. A combination of the saving on hardware purchases with a fixed service charge that bundles all software and remote IT support will not doubt be very attractive to CIOs looking to manage shrinking budgets. What will potentially get in the way of enterprise adoption is the historic reluctance of ISPs and mobile operators to offer guarantees or take risk around network outage. In a terminal-server model, network performance becomes the single thread by which the ability to carry on business hangs. In reality, even with applications running on local machines, very little work can happen when networks go down, so this risk might not be as different in the terminal-server model as it first seems. Even so, if network operators were to create new product offerings and charge a premium for a guaranteed available service (with ‘super service levels’ or other strong incentives backing up the availability promise) this might be an attractive proposition in a terminal-server world. With domestic ISPs now offering their customers credits where availability metrics are missed, this also seems like a credible direction of travel for the industry.
Looking even further ahead, ubiquitous ultra-low latency connections allow remote access to very high power computation resources for all manner of other applications. AI and machine learning techniques involving deep neural networks are computationally intensive. With modern mobile devices allowing both high quality output and high quality input from a variety of sensors, it will be fascinating to see what can be done when they are no longer dependent on low-power on board CPUs, but can be ‘live linked’ to data-centre grade computing resources.
Even if your use doesn’t change dramatically, better networks will still bring benefits for all of us. At the very least we will get to see those magic words sooner: … 100%, now playing.
For more on ubiquitous connectivity, new computing paradigms, 5G networks and the legal and regulatory issues that touch upon them, come along to DLA Piper’s European Tech Summit in London on 15th October. More details here.