dc.contributor.advisor | Yun, Heechul HY | |
dc.contributor.author | Ali, Waqar | |
dc.date.accessioned | 2022-03-19T16:04:13Z | |
dc.date.available | 2022-03-19T16:04:13Z | |
dc.date.issued | 2020-12-31 | |
dc.date.submitted | 2020 | |
dc.identifier.other | http://dissertations.umi.com/ku:17539 | |
dc.identifier.uri | http://hdl.handle.net/1808/32614 | |
dc.description.abstract | In recent years, the problem of real-time scheduling has increasingly become more important as well as more complicated. The former is due to the proliferation of safety critical systems into our day-to-day life; such as autonomous vehicles, fueled by the recent advances in artificial intelligence. The latter is caused by the increasing demand for high performance which is driving the adoption of highly integrated complex heterogeneous system-on-chip (SoC) processors to deliver the performance while meeting strict size, weight, power (SWaP) and cost constraints. Motivated by these trends, this dissertation tackles the following main question: how can we guarantee predictable real-time execution on heterogeneous multicore SoCs while preserving high utilization? The fundamental problem in preserving the determinism of the real-time system realized on a heterogeneous multicore SoC is ensuring that the worst-case execution time (WCET) of each task, measured in isolation, will stay within a reasonable bound during the actual execution of the system. The primary challenge in achieving this goal---tightly bounding task WCETs---is that the execution time of a task can be highly non-deterministic, often varying significantly depending on which tasks are co-scheduled and how they contend on various shared hardware resources in the memory hierarchy. The particular scheduling requirements (e.g., non-preemption) of the different computing resources (e.g., integrated GPU) in the heterogeneous SoC and the possible cross-contention among their workloads can also exacerbate this problem. In light of these considerations, this dissertation presents new real-time scheduling techniques for predictable and efficient scheduling of mixed criticality workloads on heterogeneous SoCs. The contributions of this dissertation include the following: 1) A novel CPU-GPU scheduling framework that ensures predictable execution of critical GPU kernels on integrated CPU-GPU platforms. 2) A novel gang scheduling framework which guarantees deterministic execution of parallel real-time tasks on the multicore CPU cluster of a heterogeneous SoC. 3) Optimal and heuristic algorithms for gang formation that increase real-time schedulability under the RT-Gang framework and their extension to incorporate scheduling on accelerators in a heterogeneous SoC. 4) Concrete evaluation results using simulated tasksets as well as real-world workloads that demonstrate the analytical and practical benefits of the proposed techniques. | |
dc.format.extent | 143 pages | |
dc.language.iso | en | |
dc.publisher | University of Kansas | |
dc.rights | Copyright held by the author. | |
dc.subject | Computer science | |
dc.subject | Computer engineering | |
dc.subject | gang scheduling | |
dc.subject | heterogeneous SoCs | |
dc.subject | integrated GPU | |
dc.subject | real-time scheduling | |
dc.subject | safety-critical | |
dc.subject | system software | |
dc.title | Deterministic Scheduling of Real-Time Tasks on Heterogeneous Multicore Platforms | |
dc.type | Dissertation | |
dc.contributor.cmtemember | Kulkarni, Prasad PK | |
dc.contributor.cmtemember | Eldin Mohamed Aly, Esam EA | |
dc.contributor.cmtemember | Davidson, Drew DD | |
dc.contributor.cmtemember | Keshmiri, Shawn SK | |
dc.thesis.degreeDiscipline | Electrical Engineering & Computer Science | |
dc.thesis.degreeLevel | Ph.D. | |
dc.identifier.orcid | https://orcid.org/0000-0002-8427-6623 | en_US |
dc.rights.accessrights | openAccess | |