Limited Preemptive Scheduling in Real-time Systems

(1)

Mälardalen University Press Dissertations No. 199

LIMITED PREEMPTIVE SCHEDULING IN REAL-TIME SYSTEMS

Abhilash Thekkilakattil 2016

School of Innovation, Design and Engineering Mälardalen University Press Dissertations

No. 199

LIMITED PREEMPTIVE SCHEDULING IN REAL-TIME SYSTEMS

Abhilash Thekkilakattil 2016

(2)

LIMITED PREEMPTIVE SCHEDULING IN REAL-TIME SYSTEMS

Abhilash Thekkilakattil

Akademisk avhandling

som för avläggande av teknologie doktorsexamen i datavetenskap vid Akademin för innovation, design och teknik kommer att offentligen försvaras

fredagen den 27 maj 2016, 13.15 i Gamma, Mälardalens högskola, Västerås. Fakultetsopponent: Associate Professor Reinder Bril, Eindhoven University of Technology

ISBN 978-91-7485-254-7 ISSN 1651-4238

(3)

Akademin för innovation, design och teknik

(4)

Abstract

Preemptive and non-preemptive scheduling paradigms typically introduce undesirable side effects when scheduling real-time tasks, mainly in the form of preemption overheads and blocking, that potentially compromise timeliness guarantees. The high preemption overheads in preemptive real-time scheduling may imply high resource utilization, often requiring significant over-provisioning, e.g., pessimistic Worst Case Execution Time (WCET) approximations. Non-preemptive scheduling, on the other hand, can be infeasible even for tasksets with very low utilization, due to the blocking on higher priority tasks, e.g., when one or more tasks have WCETs greater than the shortest deadline. Limited preemptive scheduling facilitates the reduction of both preemption related overheads as well as blocking by deferring preemptions to favorable locations in the task code.

In this thesis, we investigate the feasibility of limited preemptive scheduling of real-time tasks on uniprocessor and multiprocessor platforms. We derive schedulability tests for global limited preemptive scheduling under both Earliest Deadline First (EDF) and Fixed Priority Scheduling (FPS) paradigms. The tests are derived in the context of two major mechanisms for enforcing limited preemptions, viz., defer preemption for a specified duration (i.e., Floating Non-Preemptive Regions) and defer preemption to the next specified location in the task code (i.e., Fixed Preemption Points). Moreover, two major preemption approaches are considered, viz., wait for the lowest priority job to become preemptable (i.e., a Lazy Preemption Approach (LPA)) and preempt the first executing lower priority job that becomes preemptable (i.e., an Eager Preemption Approach (EPA)). Evaluations using synthetically generated tasksets indicate that adopting an eager preemption approach is beneficial in terms of schedulability in the context of global FPS. Further evaluations simulating different global limited preemptive scheduling algorithms expose runtime anomalies with respect to the observed number of preemptions, indicating that limited preemptive scheduling may not necessarily reduce the number of preemptions in multiprocessor systems. We then theoretically quantify the sub-optimality (the worst-case performance) of limited preemptive scheduling on uniprocessor and multiprocessor platforms using resource augmentation, e.g., processor speed-up factors to achieve optimality. Finally, we propose a sensitivity analysis based methodology to control the preemptive behavior of real-time tasks using processor speed-up, in order to satisfy multiple preemption behavior related constraints. The results presented in this thesis facilitate the analysis of limited preemptively scheduled real-time tasks on uniprocessor and multiprocessor platforms.

ISBN 978-91-7485-254-7 ISSN 1651-4238

Abstract

Preemptive and non-preemptive scheduling paradigms typically introduce undesirable side effects when scheduling real-time tasks, mainly in the form of preemption overheads and blocking, that potentially compromise timeliness guarantees. The high preemption overheads in preemptive real-time scheduling may imply high resource utilization, often requiring significant over-provisioning, e.g., pessimistic Worst Case Execution Time (WCET) approximations. Non-preemptive scheduling, on the other hand, can be infeasible even for tasksets with very low utilization, due to the blocking on higher priority tasks, e.g., when one or more tasks have WCETs greater than the shortest deadline. Limited preemptive scheduling facilitates the reduction of both preemption related overheads as well as blocking by deferring preemptions to favorable locations in the task code.

In this thesis, we investigate the feasibility of limited preemptive scheduling of real-time tasks on uniprocessor and multiprocessor platforms. We derive schedulability tests for global limited preemptive scheduling under both Earliest Deadline First (EDF) and Fixed Priority Scheduling (FPS) paradigms. The tests are derived in the context of two major mechanisms for enforcing limited preemptions, viz., defer preemption for a specified duration (i.e., Floating Non-Preemptive Regions) and defer preemption to the next specified location in the task code (i.e., Fixed Preemption Points). Moreover, two major preemption approaches are considered, viz., wait for the lowest priority job to become preemptable (i.e., a Lazy Preemption Approach (LPA)) and

(5)

Abstract

Preemptive and non-preemptive scheduling paradigms typically introduce undesirable side effects when scheduling real-time tasks, mainly in the form of preemption overheads and blocking, that potentially compromise timeliness guarantees. The high preemption overheads in preemptive real-time scheduling may imply high resource utilization, often requiring significant over-provisioning, e.g., pessimistic Worst Case Execution Time (WCET) approximations. Non-preemptive scheduling, on the other hand, can be infeasible even for tasksets with very low utilization, due to the blocking on higher priority tasks, e.g., when one or more tasks have WCETs greater than the shortest deadline. Limited preemptive scheduling facilitates the reduction of both preemption related overheads as well as blocking by deferring preemptions to favorable locations in the task code.

In this thesis, we investigate the feasibility of limited preemptive scheduling of real-time tasks on uniprocessor and multiprocessor platforms. We derive schedulability tests for global limited preemptive scheduling under both Earliest Deadline First (EDF) and Fixed Priority Scheduling (FPS) paradigms. The tests are derived in the context of two major mechanisms for enforcing limited preemptions, viz., defer preemption for a specified duration (i.e., Floating Non-Preemptive Regions) and defer preemption to the next specified location in the task code (i.e., Fixed Preemption Points). Moreover, two major preemption approaches are considered, viz., wait for the lowest priority job to become preemptable (i.e., a Lazy Preemption Approach (LPA)) and

(6)

ii

preempt the first executing lower priority job that becomes preemptable (i.e., an Eager Preemption Approach (EPA)). Evaluations using synthetically generated tasksets indicate that adopting an eager preemption approach is beneficial in terms of schedulability in the context of global FPS. Further evaluations simulating different global limited preemptive scheduling algorithms expose runtime anomalies with respect to the observed number of preemptions, indicating that limited preemptive scheduling may not necessarily reduce the number of preemptions in multiprocessor systems. We then theoretically quantify the sub-optimality (the worst-case performance) of limited preemptive scheduling on uniprocessor and multiprocessor platforms using resource augmentation, e.g., processor speed-up factors to achieve optimality. Finally, we propose a sensitivity analysis based methodology to control the preemptive behavior of real-time tasks using processor speed-up, in order to satisfy multiple preemption behavior related constraints. The results presented in this thesis facilitate the analysis of limited preemptively scheduled real-time tasks on uniprocessor and multiprocessor platforms.

iii

Faculty Opponent:

• Reinder Bril, Technische Universiteit Eindhoven, Netherlands

Examiners:

• Giorgio Buttazzo, Scuola Superiore Sant’Anna di Pisa, Italy • Gerhard Fohler, Universit¨at Kaiserslautern, Germany • Liliana Cucu-Grosjean, INRIA Paris-Rocquencourt, France

Reserve:

• Damir Isovic, M¨alardalens H¨ogskola, Sweden

Supervisors:

• Sasikumar Punnekkat, Mälardalens Högskola, Sweden • Radu Dobrin, Mälardalens Högskola, Sweden

(7)

ii

preempt the first executing lower priority job that becomes preemptable (i.e., an Eager Preemption Approach (EPA)). Evaluations using synthetically generated tasksets indicate that adopting an eager preemption approach is beneficial in terms of schedulability in the context of global FPS. Further evaluations simulating different global limited preemptive scheduling algorithms expose runtime anomalies with respect to the observed number of preemptions, indicating that limited preemptive scheduling may not necessarily reduce the number of preemptions in multiprocessor systems. We then theoretically quantify the sub-optimality (the worst-case performance) of limited preemptive scheduling on uniprocessor and multiprocessor platforms using resource augmentation, e.g., processor speed-up factors to achieve optimality. Finally, we propose a sensitivity analysis based methodology to control the preemptive behavior of real-time tasks using processor speed-up, in order to satisfy multiple preemption behavior related constraints. The results presented in this thesis facilitate the analysis of limited preemptively scheduled real-time tasks on uniprocessor and multiprocessor platforms.

iii

Faculty Opponent:

• Reinder Bril, Technische Universiteit Eindhoven, Netherlands

Examiners:

• Giorgio Buttazzo, Scuola Superiore Sant’Anna di Pisa, Italy • Gerhard Fohler, Universit¨at Kaiserslautern, Germany • Liliana Cucu-Grosjean, INRIA Paris-Rocquencourt, France

Reserve:

• Damir Isovic, M¨alardalens H¨ogskola, Sweden

Supervisors:

• Sasikumar Punnekkat, Mälardalens Högskola, Sweden • Radu Dobrin, Mälardalens Högskola, Sweden

(8)

v

“Lokah Samastah Sukhino Bhavantu.”

“May all beings everywhere be happy and free, and may the thoughts, words, and actions of my own life contribute in some way to that happiness and to that freedom for all.”

This Sanskrit verse is an expression of the universal spirit found in the ancient Indian scriptures of Vedas.

(9)

v

“Lokah Samastah Sukhino Bhavantu.”

“May all beings everywhere be happy and free, and may the thoughts, words, and actions of my own life contribute in some way to that happiness and to that freedom for all.”

This Sanskrit verse is an expression of the universal spirit found in the ancient Indian scriptures of Vedas.

(10)

Acknowledgements

The dream begins with a teacher who believes in you, who tugs and pushes and leads you to the next plateau, sometimes poking you with a sharp stick called “truth”.1

This section is my humble attempt at thanking several people who have contributed both academically and personally towards the realization of this thesis. First, let me thank my supervisors Radu Dobrin and Sasikumar Punnekkat for their unflinching support throughout the last several years that made this thesis possible. Thank you very much for believing in me, and for providing me with opportunities that positively affected the course of my studies and life. No words are enough to thank Prof. Sasi’s last minute efforts 7 years ago that gave me the opportunity to come to Sweden. Radu, it has been a privilege to be your student, and to be able to travel with you all these years. Even though I still lock my car, at least I have learned not to mistrust “not so good looking” second hand cars.

In addition to my supervisors several other people contributed towards the development of the work presented in this thesis. My sincere thanks to Sanjoy Baruah for giving me an opportunity to visit his outstanding research group at UNC Chapel Hill, as well as for providing me with a starting point into multiprocessor scheduling. Many thanks to Jim Anderson for mentoring my interests towards more

1_{This quote is widely attributed to Dan Rather, an American media personality.}

(11)

Acknowledgements

The dream begins with a teacher who believes in you, who tugs and pushes and leads you to the next plateau, sometimes poking you with a sharp stick called “truth”.1

This section is my humble attempt at thanking several people who have contributed both academically and personally towards the realization of this thesis. First, let me thank my supervisors Radu Dobrin and Sasikumar Punnekkat for their unflinching support throughout the last several years that made this thesis possible. Thank you very much for believing in me, and for providing me with opportunities that positively affected the course of my studies and life. No words are enough to thank Prof. Sasi’s last minute efforts 7 years ago that gave me the opportunity to come to Sweden. Radu, it has been a privilege to be your student, and to be able to travel with you all these years. Even though I still lock my car, at least I have learned not to mistrust “not so good looking” second hand cars.

In addition to my supervisors several other people contributed towards the development of the work presented in this thesis. My sincere thanks to Sanjoy Baruah for giving me an opportunity to visit his outstanding research group at UNC Chapel Hill, as well as for providing me with a starting point into multiprocessor scheduling. Many thanks to Jim Anderson for mentoring my interests towards more

1_{This quote is widely attributed to Dan Rather, an American media personality.}

(12)

viii

“implementable” research. My visit at UNC also gave me an

opportunity to meet Fred Brooks, who graciously accepted my request for a meeting, listened to my research, as well as gave me a few tips; many thanks to you!

I am deeply grateful to Giorgio Buttazzo for providing me with the opportunity to visit the legendary RETIS lab at Pisa. My sincere thanks to Enrico Bini and Lui Sha for allowing me to attend their courses as well as for the discussions around my research. I am also extremely thankful to Alan Burns for the feedback on mixed-criticality systems, and for the discussions around real-time systems research in general.

Rob Davis contributed to my understanding of limited preemptive scheduling by providing me with exceptional feedback. Your technical acumen and attention to details have been inspiring. Many thanks to Marko Bertogna who, as my session chair for ECRTS’14, highlighted the issue with respect to handling preemptions in multiprocessor limited preemptive scheduling, resulting in one of the papers that forms a significant contribution in this thesis

I owe special thanks to Luis Almeida for reminding me about the connection between limited preemptive scheduling and scheduling in real-time networks, and Gordana Dodig-Crnkovic for the feedback on the research methods section of this thesis. I am extremely grateful to my friends at RETIS lab and UNC Chapel Hill for all the fun during my stay. Many conference trips were wonderful mostly because of the company of Dorin, Cristian, Youcheng, Alessandra, Alessandro, Kecheng, Zhishan and Bryan– thank you for your friendship.

Of course, the wonderful colleagues and friends at IDT (who are too many to list) deserve much appreciation for making life so much fun with the coffee breaks, movies and badminton sessions. Much appreciations also go to the administrative wing at IDT, particularly Carola, Sofia and Jenny, for making the administration activities of MRTC truly “real-time”. Many thanks to Kaj for being available whenever I had some questions on safety critical systems.

Life in Sweden wouldn’t have been so much fun without the nice company of several friends in V¨aster˚as, ¨Orebro and F˚agersta. Many

ix thanks to my friends Rohan and Sreekanth for their friendship, as well as for the occasional phone calls. No amount of acknowledgements would be enough to thank Mirel who helped me with my car on numerous occasions. The otherwise boring weekends were unusually pleasant just because of the football gang (who are again too many to list). No words are enough to thank Prof. Sasi, Sunitha chechi, Kannan and Kunju for being the family away from home, and making sure that we don’t miss the homely food. Many thanks to Vipin Sir and Jessy chechi for motivating me to pursue higher studies, which has been the starting point for everything.

I would also like to thank my parents for making sure that their children get the best education, and my sister for being a nice sister of a rather forgetful brother. Also, many thanks to my in-laws for being such a friendly and caring lot. Lastly, my dear wife Meera, no words are enough to describe your love and affection, and the happiness that you have brought into my life. Thank you very much for being who you are, and for putting up with my almost endless trips.

Yes, that was a pretty long acknowledgements section! However, this thesis would remain incomplete without telling the story behind it.

Abhi, V¨aster˚as, May, 2016 The work presented in this thesis was partially supported by Vetenskapsr˚adet (The Swedish Research Council) under the project CONTESSE (2010-4276), and M¨alardalen University.

(13)