Distributed FutureGrid Clouds for Scalable Collaborative Sensor-Centric Grid Applications For AMSA TO 4 Sensor Grid Technical Interchange Meeting March 24, 2011
Methodology: To study the characteristics of the underlying distributed cloud computing infrastructure at the Network Transport messages Collaborative sensor-centric grid applications levels.
Tools: Network level -iperf and ping Transport messages level -NaradaBroker messages Collaborative sensor-centric grid applications level - SCGMMS
FutureGrid Clouds
Distributed FutureGrid Clouds: India Eucalyptus Cloud (Indiana University) Sierra Eucalyptus Cloud (UCSD) Hotel Nimbus Cloud (University of Chicago) Foxtrot Nimbus Cloud (University of Florida)
For application level measurement experiments, we ported the Grid Builder (GB) virtual GPS sensors to the FutureGrid clouds
Preliminary Results
Network Level - Throughput
Network Level – Packet Loss Rate Instance PairUnloaded Packet Loss Rate Loaded Packet Loss Rate India-Sierra0%0.33% India-Hotel0%0.67% India-Foxtrot0% Sierra-Hotel0%0.33% Sierra-Foxtrot0% Hotel-Foxtrot0%0.33%
Network Level – Round-trip Latency Due to VM
Network Level – Round-trip Latency Due to Distance
Transport Messages Level – Round-trip Latency
Application Level – Round-trip Latency
Application Level – Jitter
Future Plan Repeat current experiments to get better statistics Include scalability in the number of instances in a single cloud Study latency along the line of comparing bare metal vs VMs, product vs academic clouds, etc.
Application Level Measurement Objective: To quantify the CPU, memory and communication requirements of a broad class of naturally distributed collaborative sensor-centric grid applications on the underlying distributed cloud architectures