Efficient Ageing-Aware Techniques for Emerging Low Power Multi-Core Systems

The rise in power consumption due to the integration of more resources into a single chip poses a significant threat to performance and resource lifespan in the field of technology. As a result of this, only a fraction of the integrated resources can be powered on at any given time, hindering the primary objective of resource integration. This limitation undermines the potential benefits that could be achieved through increased resource integration. Fortunately, there are design factors that can help to increase and enhance the fraction of the integrated resources that can be powered on and utilized effectively. These factors include managing the temperature of the active nodes, the placement of dark nodes to balance the chip temperature, and the power consumption of each node. Existing techniques consider some of these factors, but do not emphasize and use the node lifetime as a variable when mapping decisions are taking place. We propose algorithms that use the node lifetime as the main variable to improve the lifetime of all the nodes alongside the temperature and the hotspot of the active resources. We approach this in two ways. Firstly, we consider a many-core system where applications arrive randomly with different behavioural patterns and use the Mean Time To Failure (MTTF) of the nodes to improve the lifetime. Secondly, we use dark nodes to alleviate the temperature and to improve the lifetime. This thesis presents the following contributions: The first approach is to use a unique dynamic rescheduling management system that takes into account node age and temperature before mapping applications. This approach consists of two different algorithms. The first algorithm improves the lifetime of nodes by migrating tasks from hot nodes to dark nodes at runtime. The second algorithm monitors and improves the lifetime of nodes by migrating tasks after an epoch. Additionally, the algorithm monitors the lifetime of two neighbouring nodes and then selects the node with the highest MTTF to improve the Lifetime. The purpose of choosing neighboring nodes is to turn one on and one off to prevent hotspot in that particular area. Additionally, the proposed method reduces temperature by ensuring an active node is surrounded by dark nodes to allow it to function at a higher frequency. Secondly, we introduce another algorithm that migrates tasks between two nodes depending on the MTTF after an epoch and also during an epoch. Then, Dynamic Voltage Frequency Scaling (DVFS) and Task Migration (TM) are used to reduce the frequency to maintain the temperature of nodes under the temperature constraint. Moreover, all the algorithms are compared with state-of-the-art of techniques in the following parameters: Temperature, MTTF and node utilisation (node utilisation is the measure of how much a node has been used - in this work, it is presented as a percentage). Additionally, the algorithms are compared with the conventional round-robin which is one of the most popular and standardized algorithm in Low Power Multi-core Systems. The first approach improves the average lifetime of all active nodes by 10%. The Second approach outperforms state-of-the-art techniques by more than 10% in MTTF.