U.S. Department of EnergyOffice of Scientific and Technical Information
RADAR: Runtime Asymmetric Data-access Driven Scientific Data Replication
Abstract
Efficient I/O on large-scale spatiotemporal scientific data requires scrutiny of both the logical layout of the data (e.g., row-major vs. column-major) and the physical layout (e.g., distribution on parallel filesystems). For increasingly complex datasets, hand optimization is a difficult matter prone to error and not scalable to the increasing heterogeneity of analysis workloads. Given these factors, we present a partial data replication system called RADAR. We capture datatype- and collective-aware I/O access patterns (indicating logical access) via MPI-IO tracing and use a combination of coarse-grained and fine-grained performance modeling to evaluate and select optimized physical data distributions for the task at hand. Unlike conventional methods, we store all replica data and metadata, along with the original untouched data, under a single file container using the object abstraction in parallel filesystems. Our system results in manyfold improvements in some commonly used subvolume decomposition access patterns.Moreover, the modeling approach can determine whether such optimizations should be undertaken in the first place.
- Authors:
- Publication Date:
- Research Org.:
- Argonne National Lab. (ANL), Argonne, IL (United States)
- Sponsoring Org.:
- USDOE Office of Science (SC)
- OSTI Identifier:
- 1498740
- DOE Contract Number:
- AC02-06CH11357
- Resource Type:
- Conference
- Resource Relation:
- Conference: 2014 International Supercomputing Conference, 06/22/14 - 06/26/14, Leipzig, Germany
- Country of Publication:
- United States
- Language:
- English
- Subject:
- access pattern; data layout; data replication; high performance computing; layout optimization
Citation Formats
Jenkins, John, Zou, Xiaocheng, Tang, Houjun, Kimpe, Dries, Ross, Robert, and Samatova, Nagiza F. RADAR: Runtime Asymmetric Data-access Driven Scientific Data Replication. United States: N. p., 2014.
Web. doi:10.1007/978-3-319-07518-1_19.
Jenkins, John, Zou, Xiaocheng, Tang, Houjun, Kimpe, Dries, Ross, Robert, & Samatova, Nagiza F. RADAR: Runtime Asymmetric Data-access Driven Scientific Data Replication. United States. https://doi.org/10.1007/978-3-319-07518-1_19
Jenkins, John, Zou, Xiaocheng, Tang, Houjun, Kimpe, Dries, Ross, Robert, and Samatova, Nagiza F. 2014.
"RADAR: Runtime Asymmetric Data-access Driven Scientific Data Replication". United States. https://doi.org/10.1007/978-3-319-07518-1_19. https://www.osti.gov/servlets/purl/1498740.
@article{osti_1498740,
title = {RADAR: Runtime Asymmetric Data-access Driven Scientific Data Replication},
author = {Jenkins, John and Zou, Xiaocheng and Tang, Houjun and Kimpe, Dries and Ross, Robert and Samatova, Nagiza F.},
abstractNote = {Efficient I/O on large-scale spatiotemporal scientific data requires scrutiny of both the logical layout of the data (e.g., row-major vs. column-major) and the physical layout (e.g., distribution on parallel filesystems). For increasingly complex datasets, hand optimization is a difficult matter prone to error and not scalable to the increasing heterogeneity of analysis workloads. Given these factors, we present a partial data replication system called RADAR. We capture datatype- and collective-aware I/O access patterns (indicating logical access) via MPI-IO tracing and use a combination of coarse-grained and fine-grained performance modeling to evaluate and select optimized physical data distributions for the task at hand. Unlike conventional methods, we store all replica data and metadata, along with the original untouched data, under a single file container using the object abstraction in parallel filesystems. Our system results in manyfold improvements in some commonly used subvolume decomposition access patterns.Moreover, the modeling approach can determine whether such optimizations should be undertaken in the first place.},
doi = {10.1007/978-3-319-07518-1_19},
url = {https://www.osti.gov/biblio/1498740},
journal = {},
number = ,
volume = ,
place = {United States},
year = {2014},
month = {1}
}