Evaluating a multicore Mixed-Criticality System implementation against a temporal isolation kernel


Mattia Bottaro and Tullio Vardanega

Presentation title

Evaluating a multicore Mixed-Criticality System implementation against a temporal isolation kernel

Authors

Mattia Bottaro and Tullio Vardanega

Institution(s)

University of Padova

Presentation type

Technical presentation

Abstract

Digital transformation causes an increasing quantity of heterogeneous software parts to be deployed in modern embedded systems. The urge to contain material cost seeks ways to integrate them in less hardware units. Safety critical systems include components must be assured to never fail, and some that may, but such that the ensuing consequences can be contained without threatening safety. Isolation is the prime response to this need, ratified in the Time and Space Partitioning (TSP) paradigm. This yields low CPU utilization, though, as ample precautionary resource margins are apportioned to components that must never fall short of resources. Alternative models have been explored to achieve higher utilization without losing safety, under the umbrella term of Mixed-Criticality Systems (MCS).

This work uses a concrete implementation of a state-of-the-art MCS solution to assess how viable the MCS premises are in practice and how they compare to the TSP solution for sustained performance. The idea originates from (1) the fact that there is no real implementation of any MCS models in the literature and (2) there is no comparison between the two approaches (MCS vs TSP) despite pursuing the same goal: the isolation between applications of different criticality. In particular, we have developed an Ada runtime environment, publicly available under GPL v3.0, supporting a MCS model and evaluated its practical limitations. Subsequently, we have compared its sustained performance against a system adhering to the TSP paradigm. Evaluating the runtime environment, we show that such an MCS model is reality-proofed and what kind of task slips out of its control. Moreover, comparing the sustained performance we show that the runtime environment achieves levels of processor utilization much higher than the TSP approach.


Additional material

  • Presentation slides: [pdf]

For more details on this presentation please click the button below: