Abstract
Modern large-scale heterogeneous computers incorporating GPUs offer impressive processing capabilities. It is desirable to fully utilize such systems for serving multiple users concurrently to visualize large data at interactive rates. However, as the disparity between data transfer speed and compute speed continues to increase in heterogeneous systems, data locality becomes crucial for performance. We present a new job scheduling design to support multi-user exploration of large data in a heterogeneous computing environment, achieving near optimal data locality and minimizing I/O overhead. The targeted application is a parallel visualization system which allows multiple users to render large volumetric data sets in both interactive mode and batch mode. We present a cost model to assess the performance of parallel volume rendering and quantify the efficiency of job scheduling. We have tested our job scheduling scheme on two heterogeneous systems with different configurations. The largest test volume data used in our study has over two billion grid points. The timing results demonstrate that our design effectively improves data locality for complex multi-user job scheduling problems, leading to better overall performance of the service.
| Original language | English (US) |
|---|---|
| Pages | 523-533 |
| Number of pages | 11 |
| DOIs | |
| State | Published - Dec 12 2012 |
| Event | 2012 IEEE International Conference on Cluster Computing, CLUSTER 2012 - Beijing, China Duration: Sep 24 2012 → Sep 28 2012 |
Other
| Other | 2012 IEEE International Conference on Cluster Computing, CLUSTER 2012 |
|---|---|
| Country | China |
| City | Beijing |
| Period | 9/24/12 → 9/28/12 |
Fingerprint
Keywords
- GPU clusters
- job scheduling
- multi-user volume rendering
- parallel volume visualizatoin
ASJC Scopus subject areas
- Software
Cite this
A job scheduling design for visualization services using GPU clusters. / Hsu, Wei Hsien; Wang, Chun Fu; Ma, Kwan-Liu; Yu, Hongfeng; Chen, Jacqueline H.
2012. 523-533 Paper presented at 2012 IEEE International Conference on Cluster Computing, CLUSTER 2012, Beijing, China.Research output: Contribution to conference › Paper
}
TY - CONF
T1 - A job scheduling design for visualization services using GPU clusters
AU - Hsu, Wei Hsien
AU - Wang, Chun Fu
AU - Ma, Kwan-Liu
AU - Yu, Hongfeng
AU - Chen, Jacqueline H.
PY - 2012/12/12
Y1 - 2012/12/12
N2 - Modern large-scale heterogeneous computers incorporating GPUs offer impressive processing capabilities. It is desirable to fully utilize such systems for serving multiple users concurrently to visualize large data at interactive rates. However, as the disparity between data transfer speed and compute speed continues to increase in heterogeneous systems, data locality becomes crucial for performance. We present a new job scheduling design to support multi-user exploration of large data in a heterogeneous computing environment, achieving near optimal data locality and minimizing I/O overhead. The targeted application is a parallel visualization system which allows multiple users to render large volumetric data sets in both interactive mode and batch mode. We present a cost model to assess the performance of parallel volume rendering and quantify the efficiency of job scheduling. We have tested our job scheduling scheme on two heterogeneous systems with different configurations. The largest test volume data used in our study has over two billion grid points. The timing results demonstrate that our design effectively improves data locality for complex multi-user job scheduling problems, leading to better overall performance of the service.
AB - Modern large-scale heterogeneous computers incorporating GPUs offer impressive processing capabilities. It is desirable to fully utilize such systems for serving multiple users concurrently to visualize large data at interactive rates. However, as the disparity between data transfer speed and compute speed continues to increase in heterogeneous systems, data locality becomes crucial for performance. We present a new job scheduling design to support multi-user exploration of large data in a heterogeneous computing environment, achieving near optimal data locality and minimizing I/O overhead. The targeted application is a parallel visualization system which allows multiple users to render large volumetric data sets in both interactive mode and batch mode. We present a cost model to assess the performance of parallel volume rendering and quantify the efficiency of job scheduling. We have tested our job scheduling scheme on two heterogeneous systems with different configurations. The largest test volume data used in our study has over two billion grid points. The timing results demonstrate that our design effectively improves data locality for complex multi-user job scheduling problems, leading to better overall performance of the service.
KW - GPU clusters
KW - job scheduling
KW - multi-user volume rendering
KW - parallel volume visualizatoin
UR - http://www.scopus.com/inward/record.url?scp=84870697204&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84870697204&partnerID=8YFLogxK
U2 - 10.1109/CLUSTER.2012.63
DO - 10.1109/CLUSTER.2012.63
M3 - Paper
AN - SCOPUS:84870697204
SP - 523
EP - 533
ER -