Loading...

I/O Workload Characterization in High-Performance Computing (HPC) Applications

Davoudabadi Farahani, Ali | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 57597 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Asadi, Hossein
  7. Abstract:
  8. In recent years, the growth of data-intensive applications has caused the storage layer to become the performance bottleneck in High-Performance Computing systems. I/O workload characterization with the help of evaluation of applications pattern provides improving the performance of the storage part of the system. Prior research have addressed a limited range of workloads and evaluated small number of storage layer parameters. In this thesis, for the first time, we provide a comprehensive study of the characterization of real High-Performance Computing applications and evaluate the I/O pattern of a wide range of workloads. After collecting complete results, we proceed to analyze them and finally, present some suggestions for improving the performance of the storage system. The results of this thesis show that finding bottlenecks and fine-tuning machine learning workloads can lead to up to 10 times decrease in runtime and up to 8 times increase in resource utilization. In OpenFOAM, because of the ability to read and write a large volume of data in a short time, utilizing powerful storage subsystems such as Burst Buffer can enhance the performance of the application. Also, in database workloads, using I/O cache accelerates queries with small record sizes
  9. Keywords:
  10. High Performance Computing ; Characterization ; Data Storage ; Input/Output Intensive Workloads ; Block Input/Output Layer ; High-Performance Computing Workloads

 Digital Object List

 Bookmark

...see more