Design Evaluation of a Performance Analysis Trace Repository
Research output: Contribution to journal › Research article › Contributed › peer-review
Contributors
Abstract
Parallel and high performance computing experts are obsessed with performance and scalability. Performance analysis and tuning are important and complex but there are a number of software tools to support this. One methodology is the detailed recording of parallel runtime behavior in event traces and their subsequent analysis. This regularly produces very large data sets with their own challenges for handling and data management. This paper evaluates the utilization of the MASi research data management service as a trace repository to store, manage, and find traces in an efficient and usable way. First, we give an introduction to trace technologies in general, metadata in OTF2 traces specifically, and the MASi research data management service. Then, the trace repository is described with its potential for both performance analysts and parallel tool developers, followed with how we implemented it using existing metadata and how it can utilized. Finally, we give an outlook on how we plan to put the repository into productive use for the benefit of researchers using traces.
Details
Original language | English |
---|---|
Pages (from-to) | 2190-2199 |
Number of pages | 10 |
Journal | Procedia Computer Science |
Volume | 108 |
Publication status | Published - 2017 |
Peer-reviewed | Yes |
External IDs
WOS | 000404959000226 |
---|---|
Scopus | 85027305000 |
ORCID | /0000-0002-5437-3887/work/154740525 |
Keywords
Keywords
- Performance Analysis, Data Repository, Metadata Extraction