Skip to content
Published: 22 Nov, 2018

Effective data processing in the decentralised cloud infrastructure of the future

NEWS Data traffic management will grow like wildfire with future self-driving cars, smart video surveillance and automated factories. In his thesis, Amardeep Mehta developed methods to optimally allocate resources to next generation applications in distributed "Mobile Edge Clouds". They are a kind of infrastructure with large centralised data centres combined with distributed computing capacity linked by 5G networks.

The rapid development of high-performance and cost-effective online units and the development of 5G networks have led to a never-before-seen growth in data traffic, which requires efficient data management. For example, manufacturers, such as Toyota and Hitachi, have predicted that the amount of data generated by self-driving cars will be several petabytes a day. Correspondingly, smart factories (so-called industry 4.0) will revolutionise the manufacturing industry.

These kinds of autonomous applications will demand very high bandwidth, low latency and minimal jitter (variation in performance). Today's centralised and remotely located data centres are incapable of providing the performance these applications require.

One way of handling these challenges is to move computing capacity from the data centres closer to the end users, which provides better performance and robustness for applications and can also reduce the costs. This paradigm, with computing capacity placed between the data centres and the applications, is usually called Mobile Edge Clouds (MECs). By optimally allocating hardware resources in the MEC infrastructure to applications, a telecom operator of a MEC infrastructure can both improve availability and performance of the applications and reduce operating costs.

However, there are several challenges to overcome. The first is determining how much computing capacity needs to be installed, and where in the network it should be placed for optimal cost and performance operation of the entire MEC infrastructure. The second, related challenge is how much capacity should be allocated to each application, and where in the network this should take place to obtain the best possible performance at the lowest possible cost. The third challenge is how a framework should be designed to simplify the development of new applications that can be run anywhere in the MEC infrastructure, ranging from the central data centres all the way out to the mobile units.

In his thesis, Amardeep Metha defines models and algorithms both as simulation tools and software frameworks for handling the challenges of resource allocation in a heterogeneous MEC environment consisting of mobile units, distributed computing nodes and centralised data centres.

"With the methods I've developed, a telecom operator can optimise the installation of a MEC infrastructure and save up to 67 per cent of the operating costs for bandwidth-intensive applications," says Amardeep Metha.

The research was conducted at the Department of Computing Science at Umeå University in cooperation with researchers at Lund University and Ericsson Research.

Amardeep Mehta was born and raised in India. He holds an engineering degree in information technology from Punjab University, Chandigarh, India, and a Master's degree in computer science from Uppsala University.

Read the thesis digitally