Application of deep learning for predicting schedules in real-time systems

Hard real-time systems are often used in safety critical systems: a task missing a deadline can be catastrophic for the system and endanger human lives. To guarantee that it meets every deadline, hard real-time systems are designed to have deterministic behavior. However, such determinism is prone to timing inference attacks. Using an analytical approach, an inference attack can be launched with a priori knowledge about the task-set. However, the advancements in deep learning opens new methods that can be used to carry out such attacks. We believe that the current state of machine learning algorithms is powerful enough to launch the attack without the complete a priori knowledge. Therefore, we propose a novel architecture that will accurately predict future occurrences of target tasks in systems using real-time scheduling algorithms. We intend to use minimal information, for instance by observing only the sequences of busy intervals and rest intervals. The architecture will: infer size of the task-set, map tasks to each time steps of busy intervals and predict future task execution.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2018

Author dc:creator

• Kim, Kyo Hyun

Contributors dc:contributor

• Mohan, Sibin

• machine learning
• real-time systems
• task prediction
• real-time scheduling

Statement dc:rights

• Copyright 2018 Kyo Hyun Kim

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/101373

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/101373