Issues of Running Large Scientific Applications in a Distributed Environment
Problem
Outline
Cactus
Sequential code
Sequential code (cont.)
Sequential behavior - cache
Parallel code
Parallel code (cont.)
Memory efficiency
Optimal ghost-zone size
Theoretical efficiency
Measured efficiency
Cluster of supercomputers
Efficiency (cluster)
Ghostzone size (cluster)
Megacomputer
Load balancing
TCP scaling
TCP - channel sharing ability
Conclusions
Email: matei@cs.uchicago.edu
Home Page: http://www.cs.uchicago.edu
Download presentation source