Technology previously used to track airplanes and space vehicles is being applied by Carleton University researchers to track the behavior of IT systems for better performance planning.
Monitoring the external behavior of applications — how busy a system is or how long responses take — can determine a performance model that essentially describes how well the system will fare over time, said Murray Woodside, a systems and computer engineering professor at Carleton University.
Actually, the performance model stemmed from the need to manage data centers supporting enterprise Web sites, and to give site visitors consistent and smooth service, he said.
The collaborative project between Woodside and Carleton University PhD student Tao Zheng, with the IBM Center for Advanced Studies (CAS), was named the CAS Project of the year. The selection was based on research excellence and potential to influence new products.
The idea behind tracking system performance, said Woodside, is based on the fact that system parameters vary depending on how many visitors a site may have at a time. “If there’s a burst of traffic, you may want to add extra Web servers or even an extra database copy.”
The model grants IT managers an inside view of applications based solely on available external behaviors. “We use the model to do ‘what ifs,'” he said. “What if we add a process, or what if we add another database, or what if we cut down on the number of requests we’re allowing the system to serve?”
Applying the space vehicle tracking technology to IT systems is admittedly “a bit of a jump”, according to Woodside, but the problem the model seeks to solve is not exactly new. The issue has been brought to the fore, he said, as corporate IT environments morphed to include Web-based systems with increasingly variable parameters.
Before the proliferation of corporate Internet sites, Web-based systems resided internally and were used primarily by employees — factors easily controlled and understood, said Woodside. “And now we have millions of potential users and nobody knows from hour to hour how many of them are going to use the system.”
Although the model was originally built from the need to better manage data centers with predominantly Web-based apps, the performance model is applicable to non-Web-based systems as well.
According to Zheng, the performance model enhances IBM’s Tivoli Intelligent Orchestrator, a data center system management tool, by performing system provisioning based on additional factors like application response time.
The tool previously determined provisioning by measuring utilization alone. “The previous [Tivoli] product was not able to deal with this kind of problem.”
By considering response times, added Woodside, IT managers can anticipate system changes and be granted extra time to deploy additional resources. Although Woodside isn’t involved in the commercialization aspect of his research, he said the model is quite applicable to other vendor system management tools like HP OpenView.
But other applications for the enterprise, he said, include managing complex data centers housing multiple applications by determining how a change to one application might affect others.
The team of researchers is also looking into applying the model to the software-testing process to help developers plan the length of the testing phase.
Besides aiding resource management, Zheng foresees the proliferation of virtualization benefiting from the ability to plan machine provisioning. “If virtual machines are a new trend in the IT world, then this technology is quite useful.”