Meeting Minutes: Team Meeting w/ Priya Time: 04/07/04 @ 6:00PM-6:30PM Where: MSE Cave Attendees: Ackley, Boyer, Fry, Wilson Purpose: Weekly Status Meeting ------------------------------ --- Presented data collected from last 2 weeks --- Priya's Comments: *) Try to turn off fault detection timeout - to confirm that 'faulty' graph has consistent spike size *) See if we can modify JBoss timeout value. *) Try large fixed size messages. *) Note: Data shows that with variable sized messages, it is difficult to find parameters to tune the system, as would otherwise be possible with fixed size messages. Any particular optimization value, such as fault-detection time, could get lost in natural variability associated with message size (such as transmission or processing time.) *) Find sweet spot for Fault-detection. *) Suggest using throughput as our metric to chase. - optimize for some minimum throughput in presence of some number of faults. *) Prepare argument for why optimizing 'x' is hard - such as "due to inherent jitter > 4ms due to message size variability. *) Consider load balancing - do we currently balance load effectively, can we imporove on this despite variable message size? - "Predictive Load Balancing" would be very interesting, but unlikely there is time for adaquate analysis. *) Upcoming Schedule - Next meeting will be last day for presenting data - Week after we will begin work on presentation material ------------------------------------ Team Meeting: Time: 04/07/04 @ 5:00PM-6:00PM Attendees: Ackley, Boyer, Fry, Wilson Where: NSH Atrium Purpose: Prepare for Meeting w/ Priya New Assignments: ---------------- AB - Run w/ Large fixed messages - Turn off fault detector (modify ReplicaMonitor) - do not increase ping period, just increase ping count - Redo "faulty variable message size" data - Redo multi-client fault-free data PW - Fix unique MsgID (clientID + timestamp) - catch/debug non-unique msgID exceptions - Modify MPClient to keep server. GA - Investigate modifying JBoss timeout value - implement round-robin load balancing - this will become prioritized load balancing. CF - Split failure/succeess time - Track message size to graphs