2 Distributing management functionality

From the structuring point of view, the main components of management systems are as follows:

• Sensors collect events occurring on managed systems. In order to achieve this, sensors monitor the changes of management-relevant parameters and emit predefined notifications. The set of all parameters is termed the Management Information Base (MIB) and may be structured in an object-oriented way.
• Sieves: They process notifications emitted by the sensors. The main tasks of a sieve consist in filtering the incoming notifications and in transforming them into information by condensing one or more notifications into a single message. In a wider sense, the task of event correlation is also done by sieves. An example for such a transformation is the generation of an alarm if an exceeding of a previously defined threshold on one or more parameters is reported by the sensors.
• A Management Algorithm whose purpose is to take decisions and to trigger actions based on the information presented by the sieves. This role might be taken either by a human expert or a knowledge-based system.
• Actors, which enforce the actions triggered by the management algorithm. This is usually done by performing modifications on one or more managed systems i.e. setting parameters.

  

Figure 1: The Management Control Cycle in platform-centric Systems

These four base components implement the management control cycle depicted in figure 1 and are distributed in systems based on a platform-centric paradigm as follows: Sensors and actors reside on the managed system; together, they form the building blocks of the agent. The managing system or "manager" consists, among other components like graphical user interfaces and databases, of sieves and management algorithms. In order to relate events and actions to a specific managed system, the managing system has to store the logical descriptions of all the managed systems it surveys. Managing system and managed system exchange their data via a communication channel termed management protocol.

While the gathering of management-relevant data and the execution of actions is done in a distributed manner, the treatment of the data remains central on the management platform. Currently available state-of-the-art management platforms like IBM NetView for AIX or HP OpenView are based on this paradigm. As a consequence, these systems are very complex and pose high requirements on the resources of the underlying hardware.

The main drawback of this centralized approach is the fact that it is unscalable because the successful operation of such a system is bound to the following factors: the number of managed systems, the number of the managed system's parameters and the bandwith of the communication channel between managing and managed system. If any of the former ones increases or if the latter decreases, the whole network becomes unmanageable. In summary, by using a platform-centric approach, the management system and the communication channel are likely to become bottlenecks.

Having the above described base components in mind, this situation can only be avoided either by distributing parts of the sieves and the management algorithms across the common run time environment or by explicitly delegating the functionality to the managed systems.