| Title | Data Provenance for Experiment Management of Scientific Applications on GPU |
| Publication Type | Conference Paper |
| Year of Publication | 2019 |
| Authors | Kim, Sejin, Oh, Jisun, Kim, Yoonhee |
| Conference Name | 2019 20th Asia-Pacific Network Operations and Management Symposium (APNOMS) |
| Keywords | application characteristics, application management, co-located applications, composability, CUDA scientific applications, data provenance, data provenance management, diverse memory usage patterns, execution patterns, experiment management, GPU, GPU environments, GPU memory usage, GPU nodes, GPU runtime environments, graphics processing units, Human Behavior, Instruction sets, Memory management, memory virtualization methods, Metrics, Monitoring, multipurpose applications, natural sciences computing, parallel architectures, pattern change, Physical Memory, Provenance, pubcrawl, Resiliency, resource configuration, Runtime, runtime configuration, runtime execution, runtime monitoring, scientific applications, scientific workflow, Task Analysis, Throughput |
| Abstract | Graphics Processing Units (GPUs) are getting popularly utilized for multi-purpose applications in order to enhance highly performed parallelism of computation. As memory virtualization methods in GPU nodes are not efficiently provided to deal with diverse memory usage patterns for these applications, the success of their execution depends on exclusive and limited use of physical memory in GPU environments. Therefore, it is important to predict a pattern change of GPU memory usage during runtime execution of an application. Data provenance extracted from application characteristics, GPU runtime environments, input, and execution patterns from runtime monitoring, is defined for supporting application management to set runtime configuration and predict an experimental result, and utilize resource with co-located applications. In this paper, we define data provenance of an application on GPUs and manage data by profiling the execution of CUDA scientific applications. Data provenance management helps to predict execution patterns of other similar experiments and plan efficient resource configuration. |
| DOI | 10.23919/APNOMS.2019.8892997 |
| Citation Key | kim_data_2019 |
Data Provenance for Experiment Management of Scientific Applications on GPU