Read an article (Never heard of Edge Computing….) this week on ATT’s presentations at their Spark Conference. Apparently, ATT is saying that the problem with AR, VR/immersive gaming, self-driving cars, drones, etc. has been two fold, lack of bandwidth and processing latency.
The long latency issue comes from having current processing for these devices being done mostly at cloud data centers, 100s of miles away from the device doing the work.
The upcoming 5G rollout should hopefully solve the bandwidth problem (for now at least) but the processing latency issue can only be dealt with by moving compute closer to where it’s needed.
A couple of weeks back I was at VMworld and one of the big announcements there was vSphere supporting 64 bit ARM processors. Pat and others talked up the coming edge processing tsunami, that will overtake IT as we know it today and bring significant benefits to everything from traffic management, to infrastructure maintenance, to better security for all, etc. Windows Server has been ported to ARM for Azure apps for a while now but I don’t know if it’s been slated for external release
The new edge
Up until this point, I had always considered edge devices as sensors and other equipment embedded in buildings, land, sea, air, machinery, etc., that provided useful, realworld information/status about their environments and when somethings gone wrong, that has to be fixed. I hadn’t really saw AR and, VR immersive gaming as an edge issue. However, drones and self-driving cars are edge devices.
AR seems to rely on smart phone levels of computation and VR today is usually tethered to a desktop PC or Mac. But to take AR and VR to the next level, processing requirements need to go up.
Self-driving cars have their own army of compute processing and sensors to deal with realtime road recognition and accident avoidance. Drones have smart phone levels of compute onboard and a nearby laptop for additional processing and control support. Not sure that edge processing requirements for these devices is increasing but I’m no expert.
But, they all need more low-latency computation to become more effective, they all require lot’s of bandwidth and some of them at least, can only perform well, if both of these requirements are solved.
CloudLets
ATT has been experimenting with neighborhood data centers, test zones or cloudlets to supply this new, low-latency processing.
These are apparently local (edge) mini-datacenters that host edge electronics gear for to \ow latency latency processing. ATT has one current test zone (or cloudlet) set up in Silicon Valley and has plans to roll out more across the US.
Up until this point, I thought edge processing would be solved by moving AI and other compute resources out to the devices themselves (see my AI processing at the edge post). Moore’s law would allow today’s compute capabilities to be embedded in low-power edge devices in a decade or so.
But why wait. If you can setup a mini-(ARM based)-data center in a neighborhood cell-phone/telephone/cable/electrical cabinet, running vSphere or Windows virtualization, with high speed networking data connections to edge devices and the cloud, you can get by with less compute processing at the edge devices, enjoy low-latency responsiveness and use less cloud resources to boot.
~~~~
Doesn’t this mean we need mini-racklets, to stack our mini cloudlets compute resources, something like 9.5″ wide and 0.5U shelving.
Just when I thought (edge) decentralization would take over compute again, cloudlets come to take it back again.
Photo Credit(s): L10000901-Edit|Guide van Nispen
Augmented Reality RFid Cup|JeanBaptisteParis
