Meeting Minutes: Team Meeting w/ Priya Time: 03/31/04 @ 6:00PM-6:30PM Where: MSE Cave Attendees: Ackley, Boyer, Fry, Wilson Purpose: Weekly Status Update ------------------------------------ -- Presented measurement data -- *) presented data including: - Fault-Free turnaround w/ one client. - Fault-Injection turnaround w/ one client. - Fault-free turnaround w/ multiple clients & servers *) Priya suggests obtaining detailed measurements of Faulty case. - Suggest pie-chart of breakdown with percentages to: time spent contacting replica timeout due to fault time spent processing - breakdown fault to prove where faulty-case can be improved. *) Andrew presented a suspected breakdown of timing data - Suggests that majority of time is spent in timeout from Fault. *) Priya suggests that we investigate if JBoss socket timeout is configurable. - If not, we may wish to consider active-replication to minimize the time spent waiting for the client to detect the fault. - Also, may consider a client that actively pings system to detect faults (at a rate faster than the JBoss socket timeout). - Suggests using "strace" on failover to see what parameters to "socket" are being used - a null timeout or a real value... *) Hard to decide load-balancing strategy until we can gather statistics, but major concepts are: - Client must cache several replica connections - Either an ordered or weighted list - Something must poll system to determine load of each replica - probably the load balancer - Client must periodically refresh it's cache with new replica loads. - probably obtain this from server (LoadBalancer) *) Priya comment: - Probably a week of coding effort left. *) Team to discuss further after class tomorrow. New Assignments: ---------------- PW - Collect detailed fault-time breakdown data AB - Revise current Fault-Free multi-client data - increase number of messages processed per client to match prior Fault-Free single-client case. CF - Investigate configurable JBoss timeout