Gautam Jayantilal Gala

• Multicore processors and Multiprocessor System-on-Chip (MPSoC) have become essential in Real-Time Systems (RTS) and Mixed-Criticality Systems (MCS) because of their additional computing capabilities that help reduce Size, Weight, and Power (SWaP), required wiring, and associated costs. In distributed systems, a single shared multicore or MPSoC node executes several applications, possibly of different criticality levels. However, there is interference between applications due to contention in shared resources such as CPU core, cache, memory, and network. Existing allocation and scheduling methods for RTS and MCS often rely on implicit assumptions of the constant availability of individual resources, especially the CPU, to provide guaranteed progress of tasks. Most existing approaches aim to resolve contention in only a specific shared resource or a set of specific shared resources. Moreover, they handle a limited number of events such as task arrivals and task completions. In distributed RTS and MCS with several nodes, each having multiple resources, if the applications, resource availability, or system configurations change, obtaining assumptions about resources becomes complicated. Thus, it is challenging to meet end-to-end constraints by considering each node, resource, or application individually. Such RTS and MCS need global resource management to coordinate and dynamically adapt system-wide allocation of resources. In addition, the resource management can dynamically adapt applications to changing availability of resources and maintains a system-wide (global) view of resources and applications. The overall aim of global resource management is twofold. Firstly, it must ensure real-time applications meet their end-to-end deadlines even in the presence of faults and changing environmental conditions. Secondly, it must provide efficient resource utilization to improve the Quality of Service (QoS) of co-executing Best-Effort (BE) (or non-critical) applications. A single fault in global resource management can render it useless. In the worst case, the resource management can make faulty decisions leading to a deadline miss in real-time applications. With the advent of Industry 4.0, cloud computing, and Internet-of-Things (IoT), it has become essential to combine stringent real-time constraints and reliability requirements with the need for an open-world assumption and ensure that the global resource management does not become an inviting target for attackers. In this dissertation, we propose a domain-independent global resource management framework for distributed RTS and MCS consisting of heterogeneous nodes based on multicore processors or MPSoC. We initially developed the framework with the French Aerospace Lab -- ONERA and Thales Research & Technology during the DREAMS project and later extended it during SECREDAS and other internal projects. Unlike previous resource management frameworks RTS and MCS, we consider both safety and security for the framework itself. To enable real-time industries to use cloud computing and enter a new market segment -- real-time operation as a cloud-based service, we propose a Real-Time-Cloud (RT-Cloud) based on global resource management for hosting RTS and MCS. Finally, we present a mixed-criticality avionics use case for evaluating the capabilities of the global resource management framework in handling permanent core failures and temporal overload condition, and a railway use case to motivate the use of RT-Cloud with global resource management.