|
 |
Auto-pilot for IT systems
| | 
There is nothing more important than quality of service for IT infrastructure. Yet, the systems are becoming more and more complex, and many companies are still relying on manual (or ad hoc) methods for managing them. Looking around various industries, IBM researchers noted that feedback control, a technique of measuring the system output and conforming it to some desired value, was already used for many complex physical systems – airplane autopilots, automobile's cruise control, computer disk-drive head positioning – but not yet for helping build better or smarter management for IT systems.
Researchers quickly discovered the two principal technical obstacles to developing the technology they envisioned: instrumentation and modeling. The team needed to create good sensors for measuring IT systems and they did not already have actuators that were easy to understand and control. And, it takes a long time to build good, robust mathematical models of behavior for even simple IT systems – and the researchers did not want to be limited to managing simple systems with this solution.
And there was another obstacle, this time from the business side. As Yixin Diao, research staff member, explained, “The business owners not only look for non-marginal improvement, but require the implementation and operation cost to be kept to a minimum.”
Getting past those roadblocks required deep understanding of the IT systems and their managing problems, and innovative adaptation of various control laws. Generic Adaptive Control is based on control theory, which designs robust and effective feedback control algorithms for managing workload and configuration of modern, on demand IT systems.
Interest in this area is growing steadily. Control theory is typically applied to physical systems and has rarely been used for software systems; in fact, the research team has helped develop the first commercial software products with a control theory-based solution embedded in it. GAC has already provided enabling technologies for Utility Throttling of DB2 v8.2 (shipped in 2004), and Adaptive Self-Tuning Memory of DB2 v9 (shipped in 2006), which is regarded as one of the four key selling features of DB2 v9. These features are important as they help deliver on the promise of autonomic computing for IBM DB2, making it significantly easier to configure and maintain databases for all market segments from small-medium businesses to IBM’s large clients. And the team continues to work within IBM to find new market opportunities, particularly in the services area.
As for the future of Generic Adaptive Control, researchers maintain that they have just scratched the surface on the potential of this technology. From a research perspective, there is much more technical progress that could be made in terms of exploring the adaptive and generic aspects of the goals. Of course, there is always the possibility of using this in more of IBM’s products. Finally, the emerging area of services offers a new area of challenges and opportunity for improving service process automation and service resource optimization.
|
