The browser you are using is not supported by this website. All versions of Internet Explorer are no longer supported, either by us or Microsoft (read more here:

Please use a modern browser to fully experience our website, such as the newest versions of Edge, Chrome, Firefox or Safari etc.

Control over the Cloud : Offloading, Elastic Computing, and Predictive Control


  • Per Skarin

Summary, in English

The thesis studies the use of cloud native software and platforms to implement critical closed loop control. It considers technologies that provide low latency and reliable wireless communication, in terms of edge clouds and massive MIMO, but also approaches industrial IoT and the services of a distributed cloud, as an extension of commercial-of-the-shelf software and systems.

First, the thesis defines the cloud control challenge, as control over the cloud and controller offloading. This is followed by a demonstration of closed loop control, using MPC, running on a testbed representing the distributed cloud.
The testbed is implemented using an IoT device, clouds, next generation wireless technology, and a distributed execution platform. Platform details are provided and feasibility of the approach is shown. Evaluation includes relocating an on-line MPC to various locations in the distributed cloud.

Offloaded control is examined next, through further evaluation of cloud native software and frameworks. This is followed by three controller designs, tailored for use with the cloud. The first controller solves MPC problems in parallel, to implement a variable horizon controller. The second is a hierarchical design, in which rate switching is used to implement constrained control, with a local and a remote mode. The third design focuses on reliability. Here, the MPC problem is extended to include recovery paths that represent a fallback mode. This is used by a control client if it experiences connectivity issues.
An implementation is detailed and examined.

In the final part of the thesis, the focus is on latency and congestion. A cloud control client can experience long and variable delays, from network and computations, and used services can become overloaded. These problems are approached by using predicted control inputs, dynamically adjusting the control frequency, and using horizontal scaling of the cloud service. Several examples are shown through simulation and on real clouds, including admitting control clients into a cluster that becomes temporarily overloaded.

Publishing year








Document type



Department of Automatic Control, Lund University


  • Control Engineering


  • Cloud
  • Control Theory
  • Model Predictive Control
  • Offloading
  • Elastic Computing
  • Networks
  • Utility Computing
  • Cloud Services
  • Internet-of-Things
  • Distributed Cloud
  • 5G Systems




  • ISSN: 0280-5316
  • ISSN: 0280-5316
  • ISBN: 978-91-8039-094-1
  • ISBN: 978-91-8039-093-4

Defence date

20 December 2021

Defence time


Defence place

Lecture hall A, building KC4, Naturvetarvägen 18, Lund. Zoom:


  • Bruno Sinopoli (Professor)