Meeting Minutes: Team Meeting w/ Priya Time: 03/27/04 @ 1:30PM-2:00PM Where: MSE Cave Attendees: Ackley, Boyer, Fry, Wilson Purpose: Weekly Status Update & Demo ------------------------------------ --- Demo Fault Tolerance to Priya --- This implementation completes the FT baseline. Also included intitial performance graphs for Fault-Free & Fault-Injector scenarios. Priya Comments (our TODO list) *) Great chart - Create pie-chart which shows breakdown of where time spent for a fail-over. - For example, measure: 1) Fault-detection time 2) Obtaining a reference to middle-tier 3) System Load - Once this is complete, during system optimizations create another pie chart shows the optimized breakdown, as it relates to the first. - In other words - Initial FD-time = 1000ms - Optimized FD-time = 120ms = 12% of original - This optimized chart may illustrate some times which cannot be optimized, yet impact the system - such as the impact of the system load. *) For HP - Measure system performance under several loads. For this part we will turn our 2D graph into a 3D, with axis as follows: ^ Execution | Time | | +-------------------> / Msg sequence # / / System Load |/_ *) Goal is so that next week, we will have enough data to implement both part II (RT) & part III (HP) - This way, we avoid re-doing RT when we implement HP. *) Be sure to remove any printf/logging that may be occuring inside the code which is being measured. This will adversely affect the performance measurement statistics. *) Consider setting FD ping rate to 1ms & measuring fail-over time. - This will either confirm/eliminate that increasing the ping-rate will lead to better fail-over. *) We have a neat feature with the ping-countdown. - Try to show how this feature helps reduce the false-positives for having to re-start a replica. - Also, try to determine the "sweet-spot" for the ping-rate & the count-down value. (Use a slow-machine & load it to show timeout.) Be sure to illustrate how to find the "sweet-spot" parameters. *) Next meeting will be regularly schedule time : Wed @ 6:00PM ------------------------------------ Team Meeting: Time: 03/20/04 @ 1:30PM-2:00PM Attendees: Ackley, Boyer, Fry, Wilson Where: 4th floor niche, WeH Purpose: Prepare for Demo to Priya Design Update: -------------- *) Modify ServerStat Bean. - Change "enabled" --> "isSchedulable" - indicates if replica is included in load-balancing algorithm. - Add "isRunnable" - indicates if replica should be started on this machine. New Assignments: ---------------- CF - Performance Graphs PW - Collect performance data - update to FINE grain on JBoss output. - Load Balancing w/ Replica State - Load Balancing w/ Multiple Clients/Servers - Clarify what "activate(server)" means GA - GRM - Stay in submenu - Fix Start/Stop All Servers - Start all "enabled" servers on startup - Revise ServerMenu: - Start ALL - Stop ALL - Enable (& Start) Replica - Disable (& Stop) Replica - Add - Remove ? - ReplicaState stats seem incorrect. - appears to be corrupted as result of Passivate/Active. AB - JBoss port change ? - Update Load Balancing Algorithm.