Networks Europe Mar-Apr 2019 | Page 14

14 EDGE COMPUTING multiple sources and providing the compute for in-depth analysis and decision-making. The Tesla on the motorway makes simple decisions in the car, while a high compute, highly-connected data centre out of town is receiving data from the vehicle and hundreds of others, analysing that data and making complex decisions for all of those cars. For example, often cited applications, such as autonomous vehicles, with their massive sensor arrays and ultra-severe latency requirements, need their intelligence to be local. It needs to be in-vehicle in order to facilitate an instant decision – braking to avoid a pedestrian. If remote processing was required, you would need a remote data centre presence every tenth of a mile, and even that would risk an ill-timed loss of connection. If there’s an edge, it’s in the autonomous vehicle itself and not in some nearby data centre at the closest 5G small cell site. If you categorise IoT applications that are truly latency sensitive, most, if not all, require lots of local intelligence. Moving forward, the role of the data centre will be in handling this new and varied complexity – joining data from Micro data centres remain vital There will be instances where micro-data centres become vital. For example, when a fire occurs, fire fighters will arrive at the scene and must create an instant intelligent edge – the at-scene command centre. The intelligent decision support must be at the scene or at the edge. Data from body-worn sensors and other items of equipment must be integrated into a common operational picture for both commanders on-site, and operators at the command centre. Since they had no previous knowledge of where the situation would occur, the edge must be instantly created. Adding to the challenge is the unknown on-site network availability and relying on a centralised location for decision support is impractical. The ultra- latency sensitive data must be processed at the edge for immediate support. On- scene data will also be sent to the central location for after-action analysis, ML and AI processing. Another reason often put forward for micro data centres at the edge is data pre- processing. Since a multitude of sensors are sending so much data upstream, you must have a location with intelligence to pre-process the data and to compress it to save on bandwidth to the main corporate data centre location. The issue with this is the link with expensive and scarce bandwidth from the sensor to the edge data centre. Once at the edge data centre, bandwidth upstream to your primary centralised metropolitan data centre should be plentiful and inexpensive. The cost savings from this approach should, therefore, be minimal. We estimate that only 10% of IoT applications and their supporting workloads require a physical presence at the edge. The remaining 90% can be sufficiently served from the existing metropolitan data centre and co-location facilities. Unless you’re in the 10%, you need to take a complete view of the costs-versus-performance trade-offs when contemplating an intelligent IoT edge strategy. In order to make 2019 the year 5G, AI and autonomous vehicles begin to realise their potential, data centre providers need to focus not on creating a new edge, but on making connectivity, compute and interconnection more seamless and more available. The task for data centres in helping their customers respond rapidly to these changes is not to re-engineer the system, but to deliver more connectivity and compute to make those complex decisions ever faster. n www.networkseuropemagazine.com