Adaptive Grid Computing Detecting and Adjusting for Dynamic Grid Conditions
Experiments with a Grid-Enabled Computational Framework Dave Angulo, Ian Foster Chuang Liu, Matei Ripeanu, Michael Russell, Lingyun Yang Distributed Systems Laboratory University of Chicago & Argonne National Laboratory Gabrielle Allen, Thomas Dramlitsch, Gerd Lanferman, Ed Seidel, Thomas Radke Max-Planck-Institut für Gravitationsphysik
Acknowledgements This work was supported in part by the NSF-funded Grid Application Development Software project under Grant No. 9975020. We are grateful to our GrADS project colleagues for discussions on the topics discussed here.
Research Goals Investigate methods and structures for efficient Grid execution via in-depth study of a demanding application, including Constructs for adapting to heterogeneity Constructs for dynamic resource acquisition Create testbed for GrADSoft components, as they emerge Investigate utility of computational frameworks as facilitator of Grid computing
Adaptation to Dynamic Grid Environments Migrate to “faster/ cheaper” system When better system discovered When requirements change When characteristics change (e.g., competition)
Adaptation to Dynamic Grid Environments Migrate to “faster/ cheaper” system When better system discovered When requirements change When characteristics change (e.g., competition)
Cactus Worm Architecture External Processes / Services Cactus Worm Server Globus Toolkit™ GRAM Cactus Worm MigrationThorn Cactus “flesh” Performance Degradation Detection Globus Toolkit™ GRISs Migration Logic Manager (Under Development) NWS App & other thorns Resource selector Globus Toolkit™ GIIS Resource Selector Client Thorn Globus Toolkit™ GRISs Resources
Migration Demonstration GIIS Running on jupitor.isi.edu Resource Selector Service Running on hamachi.cs.uchicago.edu Cluster at UIUC Cluster at UCSD Cactus Worm Server Running on amajor.cs.uiuc.edu
Migration Demonstration GIIS Resource Selector Service Cluster at UIUC Cluster at UCSD Cactus Worm Server Computational application begins run NetLogger records performance and displays visual graph
Migration Demonstration GIIS Resource Selector Service Cluster at UIUC Cluster at UCSD Cactus Worm Server Competing application on one node steals CPU cycles NetLogger
Migration Demonstration GIIS Resource Selector Service Cluster at UIUC Cluster at UCSD Cactus Worm Server Resource Selector Service contacted NetLogger
Migration Demonstration GIIS Resource Selector Service Cluster at UIUC Cluster at UCSD Cactus Worm Server resource list sent to Worm Server NetLogger
Migration Demonstration GIIS Resource Selector Service Cluster at UIUC Cluster at UCSD Cactus Worm Server Worm Server shuts down app NetLogger
Migration Demonstration GIIS Resource Selector Service Cluster at UIUC Cluster at UCSD Cactus Worm Server Worm Server starts app on new resouces NetLogger
Performance Results
Gabrielle Allen; Dave Angulo; Ian Foster; Gerd Lanfermann; Chuang Liu; Thomas Radke; Ed Seidel; John Shalf. The Cactus Worm: Experiments with Dynamic Resource Discovery and Allocation in a Grid Environment. In International Journal of High-Performance Computing Applications Volume 15, Number 4, 2001 http://people.cs.uchicago.edu/~dangulo/papers/ This presentation: http://www.cs.uchicago.edu/~dangulo/scdemo.ppt