High-performance real-time systems design from cloud to embedded edge
Authors:
Matteo Andreozzi and Girish Shirasat
Keywords:
real-time, Arm, high-performance, mixed-criticality, QoS, SOAFEE
Abstract:
"Computer Systems are rapidly evolving and moving from being designed as fixed function general purpose, real-time, high performance systems to increasingly software defined where predictability in case of resource sharing becomes a central problem to solve, more so when the workloads are mixed critical in nature. Co-location of multiple workloads on a single computer system can improve utilisation of system resources, enabling resource re-use (e. g. IO devices, hardware accelerators, etc.) and improve the efficiency of data sharing across workloads. However, co-location also comes at the cost of potential performance degradation due to interference on shared resources, and increased uncertainty which in the case of mixed critical workloads seen in automotive and industrial segments becomes a major bottleneck to address. The above is true for the great majority of use cases - ranging from automotive, industrial to high performance computing - co-location can almost always translate into maximizing compute resources utilization and minimizing costs. At the same time, resource sharing critically increases the need for predictively and deterministically managing such systems resources and is non-negotiable system attribute in a software defined design paradigm. The advent of larger integrated platforms which will run real-time workloads alongside GPOS workloads will require the ability to provision resources in a quantifiable and predictable way and the need to deliver dynamic workloads in the cloud native design paradigm associated with software defined system makes the need to provision these resource during runtime a key differentiator in next generation high performance embedded computing systems. This will become a crucial property of future computing systems to predict worst-case execution times (WCET) for their real-time workloads and to ensure smooth and responsive operation of the GPOS workloads. In this work, we'll cover the impact of shared resources interference on heterogeneous compute platforms, and we'll define the terminology and the principles which we envision will enable deterministic and predictable execution of critical and real-time applications on high performance Arm-based platforms. We will also cover system software architectures that are being envisioned in initiatives like SOAFEE (Scalable Open Architecture For Embedded Edge) to address the need of enabling mixed critical workloads and the orchestration of it from cloud to embedded edge."