From Jim Waldo - Scale and Fail

What are the metrics for scalabilty? From DARPA CoABS we have

• Complexity
• Heterogeneity
• Number of agents

• ?? - to add places where agents execute- how many platforms, network topolgy
• Mudumbai Ranganathan - - communication complexity
• Nelson Minar - the understanability of the system.
• David Kotz - how do you vary this variable?
• ?? - regularity (lattice, nearest neighbor)
• Mudumbai Ranganathan - how scalable is the location tracking? separate naming from where names are located
• David Kotz -is that a control variable or a metric? sounds like neither? how do you measure success or complexity
• Keith Marzullo - number of administrative domains

• round trip time
• completion time
• throughput

• Larry Korba - the network load imposed by the system. You would like to use the netowrk for other services
• Dwight Duego - The number of tranistions- minimize the number of messages
• Raja Pandey - relative measures. What is the context of using a mobile agent? Compare the performance of mas w/ traditional system or itself. Could be very application specific. How to reduce the amount of search to be done
• David Kotz - what metrics were different?
• Michael Wiess - The amount of overhead needed to coordinate actions - not network load, also control,
• David Kotz - metric is the same, but you break it out yes, more than just network
• Peter Doemel - measure the load that is put on the machines supporting system how the agent system interacts with the environment
• Anand Tripathi - robustness. many kinds of exceptions. stability issues. new kinds of failure modes.
• George Cybenko -some metric that captures guarenteed correctness. what confidence do you have that the system completes independent of the time.
• Pierre-Antoine Queloz - how do metrics degrade with congestion. how do we define that the system has not scaled? when is scalability not sufficient
• Daniela Rus -how is that different from expectations of throughput?
• Pierre-Antoine Queloz - because the load can be very high, but you still have more or less good performancs
• Gul Agha - more is less.... Define what trade offs are acceptable.
• Mudumbai Ranganathan - what is fundamentally different from traditional distributed systems?
• Daniela Rus - yes, these are issues in other systems. hopefully, today we will define what's different. Ma has autonomy - can move around - contribute to load balancing - how can autonomy help out?
• ?? - utilization or effectivesss?