Scheduling Strategies for the Calvin IoT Environment

Detta är en M1-uppsats från Lunds universitet/Institutionen för reglerteknik

Sammanfattning: The connection of devices to the internet is referred to as Internet of Things (IoT). By using IoT distributed network of devices will be able to communicate with each other. The management and forming of distributed applications is, however, very complex since IoT uses devices that have different capabilities and that do not have the same communication protocols. Calvin is an open source application environment developed by Ericsson that provides a distributed cloud for IoT. The distributed cloud can help to achieve low latency by using parallel processing nodes. Calvin is built upon the actor model, using the methodology dataflow programming. Using Calvin’s programming model, the different parts of the system are represented by actors that are used to isolate and abstract functionality. Actors communicate with each other by message passing, however there is a network of runtimes that handle the actual dataflow. Several runtimes are applied in every device that will be used in IoT to get metadata about the capabilities of the devices and to get more generic and reusable actors [1]. In Calvin the scheduler is responsible for data transport and for triggering actor actions on in-data. The scheduler is necessary for execution of applications for each runtime. The current scheduler is a basic Non-Preemptive (NP) scheduler with limited knowledge because it only uses local information, such as what data has arrived from other runtimes, the rules for triggering an action, and the output data that will be sent forward to other actors whether they are local or remote. The aim of the master thesis is to improve Calvin’s current scheduler. A new scheduling policy called Round Robin will be implemented and compared with the current Nonpreemptive scheduling policy with respect to latency. A dynamic sorting part will be added to Calvin’s scheduler and implemented for the two scheduling policies. The result of this thesis shows that the new strategies named New Non-Preemptive and New Round Robin are beneficial to use when most actors are not busy for every function call.

  HÄR KAN DU HÄMTA UPPSATSEN I FULLTEXT. (följ länken till nästa sida)