Electrical and Computer Engineering ETDs
Publication Date
Summer 7-7-2023
Abstract
With the complexity of high-performance computing designs continuously increasing, the importance of evaluating with simulation also grows. One of the key design aspects is the network architecture; topology and bandwidth greatly influence the overall performance and should be optimized. This work uses simulations written to run in the Structural Simulation Toolkit software framework to evaluate a variety of architecture configurations, identify the optimal design point based on expected workload, and evaluate the changes with increased scale. The results show that advanced topologies outperform legacy architectures justifying the additional design complexity; and that after a certain point increasing the bandwidth provides limited additional benefit, indicating resources should be spent improving other aspects of the design.
Keywords
high-performance computing, distributed shared memory, topology, structural simulation toolkit, optimization
Document Type
Dissertation
Language
English
Degree Name
Computer Engineering
Level of Degree
Doctoral
Department Name
Electrical and Computer Engineering
First Committee Member (Chair)
Payman Zarkesh-Ha
Second Committee Member
Jim Plusquellic
Third Committee Member
Xiang Sun
Fourth Committee Member
Patrick Bridges
Recommended Citation
Milton, Jonathan A.. "Optimizing High-Performance Computing Design: The Impacts of Bandwidth and Topology Across Workloads for Distributed Shared Memory Systems." (2023). https://digitalrepository.unm.edu/ece_etds/607