On the Performance and Scalability of Decentralized Monitoring Using Mobile Agents

Abstract

The problem of improving the performance and scalability of current monitoring systems, which generally follow a centralised and static management model is considered herein. Several alternative solutions based on Mobile Agents (MAS), specifically tailored to network monitoring are described. In particular, the key problem of computing efficiently the initial number and location of MAS is addressed, and techniques for rapid MA deployment are described. Mathematical models for the proposed MA solutions and for the common technique of centralised polling are built and a comparative analysis of their performance is carried out. Performance is expressed in terms of monitoring trafic and delay. Results on traffic show that MAS typically offer improved performance, although they do not always scale better than polling. However, results on delay demonstrate that MA solutions improve both scalability and performance under most circumstances.

On Delays in Management Frameworks: Metrics, Models and Analysis

“…In such cases, monitoring algorithms do not scale well from a delay perspective where algorithms lost the desired quality (timeliness and temporal accuracy) when achieving tasks. This aspect of delays scaling or timeliness of management tasks has been take up by Liotta et al [1] for mobile agents based management applications as an important quality parameter for critical management tasks. Chen et al [6] assume in their work of management approaches evaluation that management delays have a non uniform random variable component that fluctuates according to the size of the managed network, but no additional study has been done to describe this non uniformness.…”

“…The data we present and analyse in this paper was collected from our JMX benchmarking platform [15] running on a cluster of 100PC (I-Cluster2) 1 where nodes are connected via a gigabyte Ethernet. We built a synthetic benchmarking application based on some widely used JMX implementations (SUN Reference implementation 1.2 and MX4J 2.0).…”

“…Many approaches have been proposed for management systems, that varies from dataoriented approaches (SNMPv2, RMON), object-oriented approaches (Corba, JMX), mobile agents approaches [1], algorithmic approaches [2,3] and more recently autonomic management and control theory approaches [4]. Those approaches aim at minimizing the management overhead and costs but still accomplishing their management tasks with their desired goals.…”

“…The timeliness and the temporal accuracy aspects require that management This paper was supported in part by the IST-EMANICS Network of Excellence project. tasks and the retrieved or altered management data must be taken and fitted within a reasonable time window [1]. Thus, one fundamental performance metric of a management algorithm is the end-to-end delay required to retrieve/alter a management attribute within a group of management nodes involved in a management task.…”

A service management toolkit for active networks

“…Different distribution algorithms and strategies can be chosen in terms of efficiency, scalability, and capabilities of the EEs. E.g., [13] shows that more efficient implementations of the installing process can be achieved if the EEs are able to clone the active installation packets, which yields to an efficient multi-casting of active packets. Service types using not the same code or the same configuration parameters need a more complex, customized installation algorithm, but can also profit from multi-casting capabilities of the active network.…”