| Abstract: |
Recently, there has been an increasing trend towards executing real-time applications on multi-core platforms. However, this complicates the design problem, as applications running on
different cores can interfere due to shared resources and mediums. In this paper, we focus on thermal interference, where a given task (τ1) heats the processor, resulting in reduced service
(due to Dynamic Thermal Management (DTM)) to another task (τ2). In real-time domain, where tasks have deadline constraints, thermal interference is a substantial problem as it directly
impacts the Worst Case Execution Time (WCET) of the effected application (τ2). The problem exacerbates as we move to mixed-criticality systems, where the criticality of τ2 may be
greater than the criticality of τ1, complicating the certification process. In this paper, we propose a server based strategy (Thermal Isolation Server (TI Server)) which can be used to
avoid thermal interference of applications. We also present a heuristic to design TI Servers to meet the timing constraints of all tasks and the thermal constraints of the system. TI Servers are
time/space composable, and can be applied to a variety of task models. We also evaluate TI Servers on a hardware test-bed for validation purposes. |
