Predictability and Utilisation Trade-off in the Dynamic Management of Multiple Video Stream Decoding on Network-on-Chip Based Homogeneous Embedded Multi-cores

Guaranteed admission control decisions in embedded multi-core platforms often cause a trade-off between predictability for utilisation. The state-of-the art shows that both these objectives cannot be optimised if the workloads are dynamic and are not known a priori. Deterministic admission control approaches use worst-case response-time calculations of the tasks and flows live in the system to analytically make safe admission decisions. These tests often result in significantly under-utilised systems. Using a poor admission controller may improve system utilisation but at the cost of low-predictability. In a soft-real time system under heavy system load it is acceptable to have a few deadline misses in order to maintain relatively higher utilisation levels. This paper presents heuristic based admission control test that attempts to find a compromise between predictability and utilisation. The heuristic adjusts the estimation of subtask deadline assignment proportionally, and uses this to estimate the lateness of tasks that are admitted and live in the system. We explore the performance consequences of the proposed heuristic-based approach using an abstract simulator, and analyse its effectiveness against a deterministic admission control test under low and heavy workloads.