Date: 08 01 2002
Some other Advanced IA's Techniques used in the market
Neugents™ to Master Complex Systems
Product names are trademarks of Computer Associates International, Inc. or their respective companies.
Conventional agent technology requires someone to work out a step-by-step solution to a predefined problem and then encode discrete steps to address it. In contrast, Neugents are trained to understand a situation. By assuming the responsibility for learning how situations are reached, Neugents effectively program themselves. They can even suggest policy changes.
Neugents study environments. After Neugents become familiar with the behavior of the environment, they can identify various operating states. This information is interesting, but the key to predictive management is identifying conditions likely to anticipate transition to another state. By referencing past behavior, Neugents can pinpoint potential failures or bottlenecks (with statistical probability) before they happen, allowing a system administrator to take preventative steps. Over time this "evasive action" increases performance, reduces downtime and enhances productivity.
Genuine enterprise management is built upon a foundation of sophisticated monitoring functionality. For resources to be efficiently managed they must first be reliably monitored. Satisfying documented as well as implied, service level agreements for distributed business applications requires automatic diagnosis and response to unanticipated events.
Management solutions must actively participate in their own administration. Attempting to stem the overwhelming complexity of an almost infinite variety of hardware, software and networking configuration variables, enterprise managers now offer recursive features, such as self-tuning agents, to get the ball rolling. Although such features provide benefits, they also present an array of new challenges. However, one technology has proven particularly promising.
CA's neural agents, known as Neugents, dynamically refine, not just threshold and severity metrics, but a multi-dimensional understanding of how the attributes of IT infrastructure and application workloads interact. Neugents help reduce the complexity of the IT infrastructure and business applications to manageable proportions by providing predictive capabilities.
An event library is a collection of common events, thresholds, and Boolean state changes related to a given database or application. Definitions are either preset or derived empirically in production settings.
Event libraries provide a quick and dirty payback by accelerating installation and simplifying the configuration process. Passively accepting the supplied events, however, is a bit like making all men and women wear the same style and size of clothing. In the workaday world every implementation requires tailoring. In a typical ERP installation, for example, accepting a statistical mean suggest by the event library will cause an eruption of alarms from some servers, while others remain dormant. Nonetheless, if they do not generate too many alarms, event libraries are useful when deployed in a well-understood and stable environment.
A more advanced form of event library is an independent software program known as an agent. Some agents are endowed with sufficient adaptability that, using a variety of criteria and techniques, threshold settings can be gradually changed. Agent development has been rapid because this technology has proven malleable in distributed environments and therefore well suited to developing flexible management architectures. Adding "tunability", which really means an ability to modify thresholds ameliorates some of the shortcomings found in static event libraries.
These IA's were tested in Unicenter
The initial concept of static libraries evolved into tunable, but still essentially linear management systems. The apotheosis of the linear tuning approach even implemented in Unicenter TNG, were intelligent agents that readjust to better define alarm thresholds, or rely on policy to achieve the same end. However, this linear approach to a complex system, as embodied by the distributed enterprise IT infrastructure, is too confining for the scope and scale of the challenge at hand. The consequence of this impedance mismatch is manifested by alarm showers and escalation meltdowns, (systemic responses) which overwhelm administrative personnel. This spiraling vortex of event data overpowers the ability of even automated systems to unravel the root cause of a problem. Moreover, defining, administering, and maintaining all the filtering and correlation rules demands expensive and rare programming talent, as well as the requisite CPU processing power to ke2ep pace.
Computer Associates strongly recommends that IT organizations implement Neugents for production environments. In product form, neural agents provide the best price/performance for managing large and complex systems. Organizations have discovered that defining an endless variety of event types, or relying on multitudes of intelligent agents usually leads to alarm showers, which are exhausting, expensive and difficult to fix. By providing predictive management capability, Neugents help organizations achieve application service levels by anticipating problems and avoiding unmanageable alarm traffic as well as onerous policy administration.
The time is ripe for neural technology to be exploited in the mainstream management market and Unicenter TNG is leading the way.