Cache Sharing and Isolation Tradeoffs in Multicore Mixed-Criticality Systems

Abstract

In mixed-critical applications, tension exists between sharing and isolation with respect to hardware resources: while strong isolation might be required for highly critical tasks, somewhat permissive sharing might be reasonable for less critical tasks to improve throughput or average-case performance. In this paper, this tension is examined as it pertains to shared last-level caches (LLCs) on multicore platforms. In particular, criticality-aware optimization techniques based on linear programming are presented for allocating LLC areas in the context of the previously proposed MC 2 (mixed-criticality on multicore) framework. Experiments are also presented that show that these techniques can result in significant schedulability improvements.

Attacking the One-Out-Of-m Multicore Problem by Combining Hardware Management with Mixed-Criticality Provisioning

“…The specific number of LLC ways allocated to the Level-C/OS partition and to the per-core Level-A and -B partitions is a tunable parameter determined on a per-task-set basis using optimization techniques that were the main contribution of the precursor paper to this one [8]. These optimization techniques seek to minimize a task set's Level-C utilization while ensuring schedulability at all criticality levels.…”

“…For the schemes that support LLC isolation, we determined allocated LLC areas using our previously proposed optimization method [8], which involves solving a linear program. Under the MC 2 -FULL-ISO scheme, we divided the overall Level-C area into fourths (rounding as necessary) to give per-core areas.…”

“…To account for implementation-related overheads in our schedulability experiments, we applied several existing overhead-accounting techniques [4,8]. While a complete, formal description of these techniques is beyond the scope of this paper, in what follows, we give a high-level description of the techniques employed, and highlight the most relevant ideas.…”

“…We handled such delays similarly to our previous work [8]. However, in our earlier work, we assumed that if Level-A jobs are split into subjobs when creating a cyclic-executive dispatching table, then PETs were already defined to account for the effects of such splitting.…”

“…In contrast, we take a more holistic approach to hardware management here, and consider hardware-management tradeoffs within and among all criticality levels. The second paper [8] tackles the theoretical problem of assigning different portions of the LLC to different criticality levels by applying a combination of set-and way-based partitioning. This last paper is actually a stepping stone towards the work described herein.…”