System Software for Resource Arbitration on Future Many- Architectures
Florian Schmaus, Sebastian Maier, Tobias Langer, Jonas Rabenstein, Timo Hönig, Wolfgang Schröder‐Preikschat, Lars Bauer, Jörg Henkel · 2020
Just like every ecosystem, the computing one is subject to permanent evolution. In this paper we identify three major challenges resulting from this evolution. Those challenges stem from the hardware and application layer likewise. For one, we entered the era of many-* hardware architectures, which poses new requirements to upper layers. And with the proliferation of the computing continuum (e.g., cloud-, fog- and edge-computing), applications become more demanding and dynamic: The system needs to be able to satisfy application-intrinsic requirements and counter application-extrinsic uncertainties. As part of our contribution we present the current and ongoing research topics of our system-software stack for future many-* architectures. We further present the various mechanisms and concepts we employ within our system-software and describe how the system-software collaborates with other layers to tackle those challenges. Those concepts include a fundamentally different execution model and control-flow abstraction, allowing for massive micro-parallelism to efficiently utilize the hardware. Since the system-software research is performed as part of a collaborative research centre, we are able to approach the challenges on all layers of the technology stack and verify our solutions on an FPGA-based prototype platform. This allows us to design mechanisms in collaboration with every layer of the technology stack, that, when put together, cooperate across layer boundaries.