Related Work

Distributed network, systems and application management is an active field of research motivated - among others - by the experiences with deploying centralized management systems in large-scale corporate networks. To overcome deficiencies of centralized management systems a lot of research has been done.

Management by Delegation (MbD) [#!goye95!#,#!schoe96!#,#!schoe97!#] defines a concept for delegating functionality from managing systems to agents in order to enhance them at runtime. The decision when this delegation should happen and which kind of functionality should be transferred to the agents is taken by the managing system, thus preventing autonomous decisions by the agents. Flexible agents [#!moun97!#] rely on MbD and exhibit a certain degree of autonomy; they are able to receive event notifications from peers and can be grouped together in order to jointly achieve a task. In recent years mobile agents [#!phka98!#,#!MA98!#] which add the concept of mobility to flexible agents or MbD have been investigated. Their roaming capabilities allow them to move across networks in order to achieve specific, pre-defined tasks. However, the applicability of mobile agents is bound by security concerns; [#!grby98!#] and [#!Vigna:98!#] discuss these aspects. Mobile management agents are designed to achieve administrative tasks on systems and software; while [#!Bieszczad-Pagurek-White:98!#] discusses the advantages of applying mobile agents to management, [#!fkk99!#] presents a Java-based environment for configurable and downloadable lightweight management applications.

The following approaches for the monitoring of application response times are commonly used today: First network monitors like the Tivoli Netview Performance Monitor can be used to monitor traffic on the network. However the packets must be correlated to actual user transactions in order to calculate response times, which is quite a complex task [#!stbu98!#]. Moreover, it is very unlikely that a customer will allow a provider to place a network monitor in his corporate network. A different approach is by using active probes like done by Geyer & Weinig's GW-TEL INFRA-XS. Here an agent at the remote site actively initiates transactions and measures the response time. The problem with this solution is that it does not monitor the actual transactions of the user but some test transactions and that it increases network and server load.