WCET derivation under single core equivalence with explicit memory budget assignment
Files
Accepted manuscript
Date
2017-06-28
Authors
Mancuso, Renato
Pellizzoni, Rodolfo
Tokcan, Neriman
Caccamo, Marco
Version
Accepted manuscript
OA Version
Citation
R. Mancuso, R. Pellizzoni, N. Tokcan, M. Caccamo. 2017. "WCET Derivation Under Single Core Equivalence With Explicit Memory Budget Assignment." 29th Euromicro Conference on Real-Time Systems (ECRTS 2017). Dubrovnik, Croatia, 2017-06-27 - 2017-06-30. https://doi.org/10.4230/LIPIcs.ECRTS.2017.3
Abstract
In the last decade there has been a steady uptrend in the popularity of embedded multi-core platforms. This represents a turning point in the theory and implementation of real-time systems. From a real-time standpoint, however, the extensive sharing of hardware resources (e.g. caches, DRAM subsystem, I/O channels) represents a major source of unpredictability. Budget-based memory regulation (throttling) has been extensively studied to enforce a strict partitioning of the DRAM subsystem’s bandwidth. The common approach to analyze a task under memory bandwidth regulation is to consider the budget of the core where the task is executing, and assume the worst-case about the remaining cores' budgets. In this work, we propose a novel analysis strategy to derive the WCET of a task under memory bandwidth regulation that takes into account the exact distribution of memory budgets to cores. In this sense, the proposed analysis represents a generalization of approaches that consider (i) even budget distribution across cores; and (ii) uneven but unknown (except for the core under analysis) budget assignment. By exploiting the additional piece of information, we show that it is possible to derive a more accurate WCET estimation. Our evaluations highlight that the proposed technique can reduce overestimation by 30% in average, and up to 60%, compared to the state of the art.
Description
License
© Renato Mancuso, Rodolfo Pellizzoni, Neriman Tokcan, and Marco Caccamo; licensed under Creative Commons License CC-BY.