A new Forrester study shows that Web site visitors will leave -- and rarely return -- if a page doesn’t load within four seconds, and that puts a premium on optimizing application servers, middleware, hardware and databases for faster Web site performance.
During the expoQ webinar Maximizing Customer Experience Through WebSphere Performance Management (part of our Best Practices for WebSphere, sponsored by Candle Corp.), IBM’s Ruth Willenborg and Candle Corp’s Mary Hall keyed on customer-focused strategies and specific tactics to speed performance of WebSphere-based systems.
“WebSphere performance problems are really challenging,” Willenborg began, “because though the problem may look like something in WebSphere, the actual cause could be anywhere in your Web site. And there are so many components involved in the site your WebSphere application is running in, what starts out looking like a WebSphere performance problem may really be anywhere.”
She and Hall set out to give troubleshooting tips to help pinpoint the needle in the haystack that is the app server, middleware, hardware and databases working together.
“The first thing you have to is look at the end-user view. He doesn’t care if you’ve got WebSphere back there, or if you’re running on a Windows platform … all he cares about is the response time,” said Willenborg, who heads IBM’s WebSphere performance monitoring and benchmarking initiatives.
“Usual suspects include networks, firewalls, unrealistic testing environments and legacy back-end systems that weren’t initially designed for the high demands of Internet access.
But Willenborg shared her own Top Ten list that focuses on end-user experience, overall system and network health and specific applications. For example, analyzing CPU patterns and disk I/O can identify performance problems. Excessive memory paging and logging should also be avoided.
She also detailed how load, throughput and response times could help isolate problems. Focusing on key servlets and EJBs, minimizing HTTP sessions and using time-based write-to-manage persistence are other optimizations.
Establishing a queuing network (“bigger is not better”), caching dynamically to satisfy requests early in the cycle, co-locatiing servlets and EJBs, limiting object creation to minimize garbage collection and fixing unexpected memory leaks rounded out her list.