I read a news release (here) recently where NVidia is proposing to partition processing between on-device and cloud-located graphics hardware…here’s an excerpt:
“Kepler cloud GPU technologies shifts cloud computing into a new gear,” said Jen-Hsun Huang, NVIDIA president and chief executive officer. “The GPU has become indispensable. It is central to the experience of gamers. It is vital to digital artists realizing their imagination. It is essential for touch devices to deliver silky smooth and beautiful graphics. And now, the cloud GPU will deliver amazing experiences to those who work remotely and gamers looking to play untethered from a PC or console.”
As well as the split processing that is handled by the Silk browser on the Kindle Fire (see here), I started thinking about that “processing partitioning” strategy in relation to other aspects of computing and cloud computing in particular. My thinking is that, over the next five to seven years (at most by 2020), there will be several very important seismic shifts in computing dealing with at least four separate events: 1) user data becomes a centralized commodity that’s brokered by a few major players, 2) a new cloud-specific programming language is developed, 3) processing becomes “completely” decoupled from hardware and location, and, D) end user computing becomes based almost completely on SoC technologies (see here). The end result will be a world of data and processing independence never seen that will allow us to live in that Minority Report world. I’ll describe the events and then will describe how all of them will come together to create what I call “pervasive personal processing” or P3.
User Data
Data about you, your reading preferences, what you buy, what you watch on TV, where you shop, etc. exist in literally thousands of different locations and that’s a problem…not for you…but for merchants and the companies that support them. It’s information that must be stored and maintained and regularly refreshed for it to remain valuable, basically, what is being called “big data.” The extent of this data almost cannot be measured because it is so pervasive and relevant to everyday life. It is contained within so many services we access day in and day out and businesses are struggling to manage it. Now the argument goes that they do this, at great cost, because it is a competitive advantage to hoard that information (information is power, right?) and eventually, profits will arise from it. Um, maybe yes and maybe no but it’s extremely difficult to actually measure that “eventual” profit…so I’ll go along with “no.” Now even though big data-focused hardware and software manufacturers are attempting to alleviate these problems of scale, the businesses who house these growing petabytes…and yes, even exabytes…of data are not seeing the expected benefits—relevant to their profits—as it costs money, lots of it. This is money that is taken off the top line and definitely affects the bottom line.
Because of these imaginary profits (and the real loss), more and more companies will start outsourcing the “hoarding” of this data until the eventual state is that there are 2 or 3 big players who will act as brokers. I personally think it will be either the credit card companies or the credit rating agencies…both groups have the basic frameworks for delivering consumer profiles as a service (CPaaS) and charge for access rights. A big step toward this will be when Microsoft unleashes IDaaS (Identity as a Service) as part of their integrating Active Directory into their Azure cloud. It’ll be a hurdle for them to convince the public to trust them, but I think they will eventually prevail.
These profile brokers will start using IDaaS because then they don’t have to have separate internal identity management systems (for separate data repositories of user data) for other businesses to access their CPaaS offerings. Once this starts to gain traction you can bet that the real data mining begins on your online, and offline, habits because your loyalty card at the grocery store will be part of your profile…as will your
credit history and your public driving record and the books you get from your local library and…well, you get the picture. Once your consumer profile is centralized, all kinds of data feeds will appear because the profile brokers will pay for them. Your local government, always strapped for cash, will sell you out in an instant for some recurring monthly revenue.
Cloud-specific Programming
A programming language is an artificial language designed to communicate instructions to a machine, particularly a computer. Programming languages can be used to create programs that control the behavior of a machine and/or to express algorithms precisely but, to-date, they have been entirely encapsulated within the local machine (or in some cases the nodes of a super computer or HPC cluster which, for our purposes, really is just a large single machine). What this means is that the programs written for those systems need to know precisely where the functions will be run, what subsystems will run them, the exact syntax and context, etc. One slight error or a small lag in the response time and the whole thing could crash or, at best, run slowly or produce additional errors.
But, what if you had a computer language that understood the cloud and took into account latency, data errors and even missing data? A language that was able to partition processing amongst all kinds of different processing locations, and know that the next time, the locations may have moved? A language that could guess at the best place to process (i.e. lowest latency, highest cache hit rate, etc.) but then change its mind as conditions change?
That language would allow you to specify a type of processing and then actively seek the best place for that processing to happen based on many different details…processing intensity, floating point, entire algorithm or proportional, subset or superset…and fully understand that, in some cases, it will have to make educated guesses about what the returned data will be (in case of unexpected latency). It will also have to know that the data to be processed may exist in a thousand different locations such as the CPaaS providers, government feeds, or other providers for specific data types. It will also be able to adapt its processing to the available processing locations such that it elegantly deprecates functionality…maybe based on a probability factor included in the language that records variables over time and uses that to guess where it will be next and line up the processing needed beforehand. The possibilities are endless, but not impossible…which leads to…
Decoupled Processing and SoC
As can be seen by the efforts NVidia is making is this area, it will soon be that the processing of data will become completely decoupled from where that data lives or is used. What this is and how it will be done will rely on other events (see previous section) but the bottom line is that once it is decoupled, a whole new class of device will appear, in both static and mobile versions, that will be based on System on a Chip (SoC) which will allow deep processing density with very, very low power consumption. These devices will support multiple code sets across hundreds of cores and be able to intelligently communicate their capabilities in real time to distributed processing services that request their local processing services…whether over Wi-Fi, Bluetooth, IrDA, GSM, CDMA, or whatever comes next, the devices themselves will make the choice based on best use of bandwidth, processing request, location, etc. These devices will take full advantage of the cloud specific computing languages to distribute processing across dozens and possibly hundreds of processing locations and will hold almost no data because they don’t have to, everything exists someplace else in the cloud. In some cases these devices will be very small, the size of a thin watch for example, but they will be able to process the equivalent of what a super computer can do because they don’t do all of the processing, only what makes sense for the location and capabilities, etc.
These decoupled processing units, Pervasive Personal Processing or P3 units, will allow you to walk up to any workstation or monitor or TV set…anywhere in the world…and basically conduct your business as if you were sitting in from of your home computer. All of you data, your photos, your documents, and your personal files will be instantly available in whatever way that you prefer. All of your history for whatever services you use, online and offline, will be directly accessible. The memo you left off writing that morning in the Houston office will be right where you left it, on that screen you just walked up to in the hotel lobby in Tokyo the next day, with the cursor blinking in the middle of the word you stopped on.
Welcome to Minority Report.