singhalshubh / Thermal-Aware-Data-Freshness-Formulation-and-Evaluation

Thermal-Aware Data Freshness: Formulation and Evaluation

This project has been done in collaboration with Dr. Deepak Gangadhar at International Institute of Information Technology Hyderabad in the capacity of Research Assistant.

Overview of Project

In safety-critical systems, one can find several applications such as Adaptive Cruise Control (ACC) with task chains, whereby a task produces output data that is consumed by another task in the chain. Typically, there is a freshness requirement for the data consumed by a task for the data to be valid. Working with invalid data can put the system at serious risk. Similarly, the reliable operation of the applications on the processor requires the processor temperature to remain within the allowed peak temperature. There isn't any other work in literature that presents an analysis framework to derive periods of a task set considering both end-to-end data freshness and thermal constraints. In this paper, we attempt to fill this gap systematically by presenting a thermal-aware data freshness analysis framework considering a 2-task scenario and then extend it to an n-task scenario. We perform evaluations for schedulability based on the task periods derived using the thermal-aware data freshness formulation proposed in this work.

How to run the code?

• Description of files

• it.cpp is a C++ code, performing EDF scheduling of real-time task-chains (Adapted from MiBench) given the temperature constraints and data freshness bounds.
• optimization.m in MATLAB, requires the values of idle time and m(splitting parts) for all the tasks for various temperature ranges along with the data freshness for every task chain fixed at 30% of the maximum freshness bound, as the input where we have considered BCET of every task to be exactly half of their WCET. The output are the periods of the tasks.
• aut.sh contains the sample task chains for the simulation.

• Running the code
  • ./aut.sh will generate the sequence in which EDF scheduling occurs and the total freshness obtained.
  • You can modify the task chains, where format is for a task in aut.sh.

Results

Schedulability Percentage and Utilization at various temperature ranges for different freshness bounds. (a)80%,(b)50%, (c)30%, and (d)12%

Schedulablilty with Different Chain Lengths


Schedulablilty with Differing Chain Lengths undervarious temperature constraints

For any queries feel free to reach out to me at [e-mail]. This research work is currently in progress(Focusing on comments from RTAS 2021).