We are attempting to productionalize our Xcelsious Connected Dashboard (webservice based). Following the industry standard of 10% concurrency model, we attempted to stress test the dashboard for 15virtual users via loadrunner. The dashboard has 2 webservice calls that are fired refresh on load. Data retrieved from sqlserver. We are finding heavy database server CPU utilization during the initial load of the dashboard. Loadrunner was scripted to open 15 URL (ip spoofing) at the same instant. After a pause of 20sec, loadrunner is scripted to fire single sql calls (simulating user making change on the dashboard to fire 1 webservice method).
WHat we find is that in the initial loading, when all 15 users attempt to open URL, there is a dramatic spike in CPU usage (90%) which then drops dramaticaly to 3-8% on steady state.
Are there any suggested solution that will mitigate the initial spike? We are considering some option such as introducing a degree of millisecond delay with randomness (10-20 milliseconds) before the refresh on load fires for each user when initializing. We assume this would solve the prevent overloading the server at the same instant. No query on the app takes more than 1 second to return data. So this is not a DB performance issue.
Any suggestions?
B
buzzer75 (BOB member since 2006-11-08)
(BOB member since 2006-10-23)