Exhaustion Dominated Performance : Methodology, Tools and Empirical Experiments
Problem with application sensitivity to insufficient resources in High Performance Computing (HPC) clusters has been a longstanding issue of concern for industries. In this thesis we propose a method based on a black box approach for characterization and analysis of engineering simulation applications with respect to available hardware resources in the situation of resource depletion. The basis of this hypothesis is the existence of one dominating bottleneck at any given moment during execution time. This method suggests that engineering simulation application’s behavior to exhaustion can be explained as an approximately linear dependency of execution time for the problem at hand.
Verification of this hypothesis required accurate non-intrusive measurement, precise methods and tools for individual depletion of hardware resources. The results of these efforts were methods for depletion of available bandwidth, available RAM memory, processor capacity of compute nodes and practical guidelines for recognition of bottlenecks.
The method of bandwidth depletion succeeded to decrease the available bandwidth, by generating stable and robust network traffic, with only 0.5% deviation from the desired target with a maximum variance of approximately 1%.
Furthermore, the method for depletion of processor capacity resulted to a successful to generate artificial CPU load and occupy as well as 90% of the processor capacity. The verification for this tool confirmed the accuracy of 0.11% median deviation. The error range for attempts for a target load lay between 0.00% at minimum and 1.04% at maximum.
Our experiments with HPL and Fluent whilst exhausting the RAM memory and the available bandwidth of the computer cluster, confirmed that our proposed method for analysis and recognition of bottlenecks is fully applicable and a viable option for characterization in this domain.
This dissertation MIGHT be available in PDF-format. Check this page to see if it is available for download.