Software Performance Monitoring Gets Heuristic

An application with crummy performance isn't much of an application at all. A new way of monitoring software performance dispenses with the fixed rules typically used to gauge how well apps work, checking software instead against performance models it builds on its own.

This method, called a heuristic approach, builds a model of how an application component can be expected to perform based on actual tests of the software. That model matches up against how the application is working in production. Acsera Corp.'s Acsera Manager uses a "bench test" process that runs Java components and produces realistic information on how they can be expected to perform in action. The information is captured in a model, which is then enlarged to include how it will interact with related components.

"You get a level of intelligence around the application that you didn't have before," says Scott Metzger, CTO of TrueCredit, a subsidiary of credit information firm TransUnion that's using the tool.

Application performance management has become more critical as businesses pay closer attention to how well software infrastructure interacts with customers and partners. When it comes to company Web sites, for example, prompt application responses are an outright necessity.

Acsera's technique is based on a measure that includes real-world knowledge specific to each component, including what might go wrong with it. When the system spots an application slowdown, it can identify which components are behind the slowdown and apply background information to guide an IT troubleshooter on what might be wrong.

TrueCredit previously assembled teams of five or six specialists to figure out what might be slowing applications, Metzger says, because it took several people to interpret actions captured in different server logs. Now one or two people use the Acsera tool to identify a problem and resolve it in 30% less time.

Conventional ways of monitoring software involve applying inflexible rules to performance data. One method consists of putting software agents on the application server to watch operations and report back to a management system. A lightweight method uses sniffers that sit on the network, capturing response times of messages flowing to and from applications. Both take the information gained and apply metrics to provide an idea of performance levels.