ISEP - DM - Engenharia de Sistemas Computacionais Críticos
Permanent URI for this collection
Browse
Browsing ISEP - DM - Engenharia de Sistemas Computacionais Críticos by advisor "Sousa, Paulo Manuel Baltarejo de"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Space is so monotonic: Introducing dynamic schedulers to satellite softwarePublication . PASCHOALETTO, ALEXANDER PINHEIRO; Sousa, Paulo Manuel Baltarejo deThe space industry has seen a trend shift in the recent decades by a handful of perspectives. The increase in competition via the introduction of more participants (both state-related and private), the expansion in mission scopes from simple research and defense to others such as internet service and even tourism, and the growing computational demand to handle these new missions are a few of them. On this scenario, satellites come as a fundamental element in most use cases. Just as any spacecraft, they need to be developed to withstand the harsh physical conditions of space, which imply techniques such as radiation hardening in some components to operate reliably, and are deployed in environments where maintenance is complicated. They are also expected to work autonomously for years, even decades. Given these and other challenges, satellites traditionally have a long and expensive development phase, and are slow-paced when it comes to incorporating recent technologies. For the on-board computers that go within them, single-core processors of outdated architectures, fixed-priority schedulers and low CPU workloads are dominant. This paradigm works for now, but may not in the years to come as trends such as Artificial Intelligence (AI) and real-time video streaming may also reach the space domain. The lastest iterations of space-oriented software do tackle the issue of development costs by introducing a greater code re-usability across missions, but little seems to be done regarding the software performance itself. In this context, this Thesis aims at bringing modern software paradigms into play by introducing the support of three widely known schedulers - Rate Monotonic (RM), Earliest Deadline First (EDF), and Constant Bandwidth Server (CBS) - into KARVEL, a space-oriented software originally developed by Critical Software. We evaluate their performance, advantages and shortcomings in both synthetic (by emulation work as busy-wait routines) and real-world workloads (by deploying it into a robot), and demonstrate that indeed dynamic algorithms such as EDF are capable of outperforming RM even on overloaded scenarios.