Intelligent Data Centres Issue 12 | Page 18

DATA CENTRE PREDICTIONS Meet Edge Computing in theory data can be stored, accessed and then uploaded to the cloud when accessible, while data resides on the edge of the device’s network. This feature greatly benefits AI devices, such as smartphones and self-driving cars which don’t always have access to the cloud due to network availability or bandwidth but are reliant on data processing to make decisions. In a world that is increasingly data-driven, much of that information is generated outside of the traditional data centre. This is Edge Computing: the processing of data outside of that traditional data centre and typically on the edge of a network onsite. Infrastructure at the edge, despite its small hardware footprint, is able to collect, process and reduce vast quantities of data so that it can be uploaded to a centralised data centre or the cloud. Instead of sending data across long routes, this allows for data to be processed and reacted upon closer to the point of creation. Many use cases, such as self-driving cars, quick service restaurants, grocery shops and industrial settings like energy plants and mines, have found Edge Computing to be key in their implementation. This said, there are still improvements to be made in how effectively information captured at the edge is used. Since AI is still in its infancy, it requires an incredible amount of resources in order to train its models. For these training purposes, Edge Computing is best suited to allow information and telemetry to flow into the cloud for deep analysis, and models that are trained in the cloud should then be deployed back to the edge. Cloud and data centres will always be the best resources for model creation. And now in practice Cerebras, a next-generation silicon chip company, just introduced its new ‘Wafer Scale Engine’ which is designed specifically for the training of AI models. With 1.2 trillion transistors and 400,000 processing cores, the new chip is phenomenally fast. However, all of this consumes a huge amount of power, which means it isn’t viable for most Edge deployments. Data lakes can be created and better utilised by organisations when consolidating Edge Computing workloads using HCI. Once data is in a data lake, it’s available to all applications for analysis. On top of this, Machine Learning can 18 Issue 12 The combination of HCI and Edge Computing also provides reduced form Phil White, CTO, Scale Computing provide new insights using shared data from different devices and applications. HCI creates an ease of use by combining servers, storage and networking all in one box. This eliminates many of the challenges of configuration and networking that come with Edge Computing. Additionally, platforms can integrate management for hundreds or thousands of Edge devices in different geographical locations all with different types of networks and interfaces. These allow for much of the complexity to be avoided, which significantly reduces operational expenses. WITH THE HELP OF HCI AND EDGE COMPUTING, ORGANISATIONS CAN HARNESS AI TOOLS FOR SMARTER DECISION-MAKING. How does AI benefit from HCI and Edge Computing? With the introduction of smart home devices, wearable technology and self- driving cars, AI is becoming much more common and is only set to grow, with an estimated 80% of devices having some sort of an AI feature by 2022. Most AI technology relies on the cloud: it makes decisions based on the collection of data stored in the cloud it is accessing. However, since the data has to travel to data centres and then back to the device, this can cause latency. Latency is especially problematic for technologies such as self-driving cars, which cannot wait for the round-trip of data to know when to brake, or how fast to travel. A key benefit of Edge Computing for AI is that necessary data would live locally to the device, which reduces latency. New www.intelligentdatacentres.com