When it comes to data centres today, we have a shiny new list of terminology; hyperscale, enterprise, co-lo, cloud and now we have ‘edge’ as the latest buzz word in the industry. James Rix, associate director for data centres at Arcadis explores how the edge will impact energy efficiency.
Edge data centres range in size from 150kW up to 1MW and the intent behind them is to bring the compute, but not storage, closer to the end user, thus reducing latency in applications – good news for those who game or trade on markets at home.
One of the primary drivers of this is not ‘World of Warcraft’ or the like but rather the move to 5G technology. Autonomous vehicles will require the transmission of many gb’s and tb’s of data.
The speed of response of these data requests will be all important. This needs to happen close to the end user, however, the storage of this data can and will only happen at a large regional or core facility further away.
An edge data centre can be ‘local’, (small foot-print) for first line processing, analysis and routing; ‘near’ or ‘regional’ to connect, integrate and re-route; and those far away ‘core’, such as hyperscale facilities.
However it is spun or renamed, a data centre is still a utility hungry shed.
Yes, size relates to power draw, although with rack densities continuing to increase, what was a common point of 2-3kW per rack over the last eight to ten years, is now a projected 14kW per rack post 2020 and 5G implementation.
The energy effect of edge, of whatever size, will be to somewhat decentralise data centres from energy/data parks/hyperscale to a neighbourhood near you.
Concerns around the draw from local power grids notwithstanding the security and feasibility/acceptance of this is still in relatively early stages. Communities may not want to see or have a ‘data centre in a box’ in their local area with the inherent risk of noise, vandalism etc.
A domestic three bed house will draw a max of 15-18kW therefore, a 150kW local edge DC will equate out to approximately eight additional houses for power draw.
Could the grid stand this, the draw is exponential on the size of the edge DC. A town of 30-40,000 inhabitants may require several of such local edge facilities, depending on use case or services being provided such as for connected or autonomous vehicles.
None of this goes to reduce energy consumption. Currently it would not be feasible to power an edge DC with a wind turbine as such devices currently only generate power 40% of the time, (up to 60% inefficiency) so a back-up or supplement would be needed to sustain a stable supply, either a storage battery, generator or utility supply. Notwithstanding the need for a UPS of some description within the circuit.
More work needs to be undertaken now to ensure that an edge DC will have a limited impact on a local utility supply grid and not just deposited where a supplier wants and plugged in to a grid supply.
By and large ‘local’ edge DC’s will be run as ‘dark’ facilities; this means no staff on-site to look after this on a full-time basis.
Operators need to have a system in place that will monitor and report automatically on the behaviour of the facility and be able to autonomously perform a range of functions to protect the integrity of the system and services. The machine learning capability is well on the way towards that, this is not artificial intelligence, however.
It must be very clearly understood that edge computing only brings the interaction with the server to a more localised, closer to end point, location. It reduces latency (the time of the response from a server) important in both on-line gaming or trading.
It does not take away the fact that an edge DC needs to communicate with ‘the cloud’/’regional’ or ‘core’, which holds the basis for higher computing power.
In other words, edge DC’s still communicate back to a central data centre and that is where they draw down and store their information. This does not reduce the amount of energy that DC’s take, but rather spreads it around to more diverse locations.