CA Unveils Unicenter TND Beta
Proactive network management is the holy grail of monitoring system -- each vendor hopes to create software sophisticated enough to reliably predict problems before they occur. In the quest to create the ultimate performance model, Computer Associates International Inc. (CA,
www.ca.com) added a weapon to its arsenal -- time.
CA recently released the beta of its latest monitoring tool, Unicenter TND. TND is an abbreviation for the The Next Dimension, pointing to the implementation of time-based reporting of network systems.
The beta release lets administrators interested in deploying Unicenter TND evaluate the product on test systems before the product becomes available. While it isn’t a finalized version, the beta contains most of the functionality planned for the full release. Allan Andersen, vice president of business management at CA, says it is an extensive beta.
In an ideal world, a network would experience consistent demand throughout the workday. In real life, networks frequently experience peaks and valleys in demand and performance.
Traditional monitoring tools employ thresholds or event-based monitoring. When the monitoring system is configured, acceptable levels of demand or performance are set. If a threshold is exceeded, a monitoring event occurs alerting the administrator of a potential problem.
This approach has two glitches when it comes to smooth network management: It can produce false alarms, desensitizing administrators, and it can only detect problems after they occur. Steve Foote, president of the Enswerf Group, says monitoring tool vendors are aiming to "figure out how to put out fires before they start." The more proactive a tool’s monitoring, the more productive the network.
With the addition of time to monitoring reports, administrators can tell when increases in use generate normal spikes, and when out-of-the-ordinary spikes threaten the stability of machines or networks. Pure event-based monitoring is often unable to accommodate a routine set of spikes.
Networks frequently experience escalating use. Time-based tools can analyze past trends and detect when failures are imminent, allowing administrators to get a feel for future network performance.
Like past versions of Unicenter, Unicenter TND employs remote agents, dubbed neugents, on all monitored devices. While agents in some monitoring systems simply relate performance data to the central monitoring station, neugents convey information that is relative to stable system performance. "It creates a multidimensional vector which is much smaller than the number itself."
The central monitoring unit determines if the vector is similar to normal machine use, or if it is wildly unstable. Rather than setting thresholds for the network, it uses a qualitative sense of normal or abnormal.
A neugent’s demand on a machine’s processor is minimal, and vector-based information requires less bandwidth and fewer cycles on the central monitor, making the system more scalable. Furthermore, the vector information does not require thresholds on individual machines.
Monitoring software is typically used by large networks with numerous servers. Event-based monitoring requires the thresholds for each machine be set individually, which can be taxing on the central monitoring station. "In a thousand server network, each one of those servers has a thousand metrics," Foote says. This would create over 10 million rules for the monitor to track.
Unicenter’s emphasis on recognizing stable states uses a different kind of intelligence than event-based monitoring. "It uses pattern recognition to detect which parts of the vectors are changing," Foote says, contrasting it with filtering data through a multitude of rules.
It takes two to three weeks for the monitoring center to establish what are stable and unstable use patterns. "There’s still time required to customize good states and bad states," Foote says, but he believes installation will be less labor intensive.
Earlier versions of Unicenter used a technology-based approach to reporting, treating networks as pure number crunchers and data transmitters. Unicenter TND uses metaphors to relate network data to how a network is used. Two new features to Unicenter are an interface, which CA dubs a portal, and data related to applications on the network.
The portal interface is Web-based, customized by users to reflect which machines are used for what. Through the portal, users can drill down to look at specific systems.
Unicenter TND also looks at how applications affect network performance, reporting on and adjusting for their use.
Unicenter’s stability-based approach to monitoring is expected to benefit e-business and dot-com companies that have wildly fluctuating networks. Dot-com firms often start small, but quickly need scalable solutions. Since Unicenter doesn’t require thresholds to be set uniquely for each machine, new equipment can be integrated into the monitor with a minimum of fiddling.