TY - GEN
T1 - A multi-level monitoring framework for stream-based coordination programs
AU - Nguyen, Vu Thien Nga
AU - Kirner, Raimund
AU - Penczek, Frank
N1 - Vu Thien Nga Nguyen, Raimund Kirner, and Frank Penczek, 'A multi-level monitoring framework for stream-based coordination programs', paper presented at the 12th International Conference on Algorithms and Architectures for Parallel Processing. Fukuoka, Japan, 4-7 September 2012.
PY - 2012
Y1 - 2012
N2 - Stream-based Coordination is a promising approach to execute programs on parallel hardware such as multi-core systems. It allows to reuse sequential code at component level and to extend such code with concurrency-handling at the coordination level. In this paper we identify the monitoring information required to enable the calculation of performance metrics, automatic load balancing, and bottleneck detection. The monitoring information is obtained by implicitly instrumenting multiple levels: the runtime system and the operating system. We evaluate the monitoring overhead caused by different use cases on S-Net as it is a challenging monitoring benchmark with a flexible and fully asynchronous execution model, including dynamic mapping and scheduling policies. The evaluation shows that in most cases the monitoring causes a negligible overhead of less than five percent.
AB - Stream-based Coordination is a promising approach to execute programs on parallel hardware such as multi-core systems. It allows to reuse sequential code at component level and to extend such code with concurrency-handling at the coordination level. In this paper we identify the monitoring information required to enable the calculation of performance metrics, automatic load balancing, and bottleneck detection. The monitoring information is obtained by implicitly instrumenting multiple levels: the runtime system and the operating system. We evaluate the monitoring overhead caused by different use cases on S-Net as it is a challenging monitoring benchmark with a flexible and fully asynchronous execution model, including dynamic mapping and scheduling policies. The evaluation shows that in most cases the monitoring causes a negligible overhead of less than five percent.
UR - http://www.scopus.com/inward/record.url?scp=84866677500&partnerID=8YFLogxK
U2 - 10.1007/978-3-642-33078-0_7
DO - 10.1007/978-3-642-33078-0_7
M3 - Conference contribution
AN - SCOPUS:84866677500
SN - 9783642330773
VL - 7439 LNCS
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 83
EP - 98
BT - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
T2 - 12th International Conference on Algorithms and Architectures for Parallel Processing, ICA3PP 2012
Y2 - 4 September 2012 through 7 September 2012
ER -