Multilevel Scheduling of Computations on Parallel Large-scale Systems

Eleliemy, Ahmed Hamdy Mohamed. Multilevel Scheduling of Computations on Parallel Large-scale Systems. 2021, Doctoral Thesis, University of Basel, Faculty of Science.


Official URL: https://edoc.unibas.ch/82695/

Downloads: Statistics Overview


Computational scientists are eager to utilize computing resources to execute their applications to advance their understanding of various complex phenomena. This eagerness drives the rapid technological development in high performance computing (HPC). Modern HPC systems exhibit rapid growth in the number of cores per computing node and the number of computing nodes per system. As such, modern HPC systems offer additional levels of hardware parallelism at the core, node, and system levels. Each level requires and employs techniques for appropriate scheduling of the computational work at the respective level. These scheduling techniques work separately without coordination, and each technique is designed to achieve specific performance targets. Currently, the absence of coordination between schedulers at different levels is an open research problem. In many cases, independent scheduling decisions degrade applications’ performance and signify inefficient resources’ usage of contemporary HPC systems. To solve this problem, we formulate the following research question: How can the multilevel parallelism of a modern HPC system be exploited through scheduling to improve the performance of computationally-intensive applications and to enhance the utilization of HPC resources? Understanding the relation between the different scheduling levels is crucial for solving the aforementioned research question. However, it is challenged by (1) the absence of methods, models, and tools that allow examining and analyzing the interaction and the mutual impact of these scheduling levels and (2) the different nature and performance targets of each of these scheduling levels. This doctoral dissertation addresses these challenges in the context of two specific scheduling classes: queuing-based job scheduling at the batch-level and dynamic loop self-scheduling (DLS) at the application-level. We propose and evaluate a multilevel scheduling (MLS) prototype that solves the problem by bridging the schedulers at these scheduling levels. The MLS prototype aims to decrease applications’ execution time and increase system utilization. It employs two novel scheduling approaches that have been introduced by this doctoral dissertation: (1) the distributed chunk-calculation approach (DCA) and (2) the resourceful coordination approach (RCA) to achieve performance targets. At the application-level, DCA addresses the scalability challenges associated with existing DLS implementation approaches while maintaining a global scheduling overview that is important to achieve global optimal scheduling decisions. We apply DCA to several DLS techniques, and we show how it benefits applications’ execution time (the first goal of the MLS prototype). At the batch-level, RCA enables application schedulers to share their allocated but idle computing resources with other applications through a batch system. The significance of RCA is that it leverages and combines the advantages of node sharing and dynamic resource and job management. It offers an efficient resource sharing (of idle resources only) and avoids shrink and expansion operations on the application side. RCA allows batch systems to reassign computing resources once they become free (the second goal of the MLS prototype). By employing DCA and RCA, the MLS prototype answers the research question and shows a creative and useful way of exploiting the multilevel parallelism of modern HPC systems through scheduling. This doctoral dissertation advances the state-of-the-art by demonstrating the usefulness and the performance potential of coordinated scheduling decisions at different levels. We also designed and implemented a set of methods and tools, which we make available for the community to analyze the mutual impact of decisions at different levels of scheduling.
Advisors:Ciorba, Florina M.
Committee Members:Schuldt, Heiko and Nagel, Wolfgang E.
Faculties and Departments:05 Faculty of Science > Departement Mathematik und Informatik > Informatik > High Performance Computing (Ciorba)
UniBasel Contributors:Eleliemy, Ahmed Hamdy Mohamed and Ciorba, Florina M. and Schuldt, Heiko
Item Type:Thesis
Thesis Subtype:Doctoral Thesis
Thesis no:14096
Thesis status:Complete
Number of Pages:xix, 130
Identification Number:
  • urn: urn:nbn:ch:bel-bau-diss140961
edoc DOI:
Last Modified:15 Jun 2021 09:00
Deposited On:07 Jun 2021 11:10

Repository Staff Only: item control page