Robust Self-Monitoring in Trusted Desktop Grids for Self-Configuration at Runtime

Abstract

A Trusted Desktop Grid System (TDG) is a platform for autonomous agents to share their computing resources based on trust relationships. Thereby, agents that only use the system without a fair participation are considered as malicious. Typically, the effects of active malicious agents and high-load situations of the TDG are similar -calling for appropriate approaches to distinguish them and, thus, allowing for counter measures to attacks. In this paper, we investigate the effect of high load to our measurements and present a concept for filtering our metrics. The evaluation demonstrated that we can normalise our metrics under high load to detect attacks with a high certainty using system-wide metrics.

Self-improving system integration: Mastering continuous change

“…In the context of computational trust, a concept for establishing explicit communities of mutually trusted subsystems and isolating malicious or faulty subsystems has been presented [95]. Based on this initial concept, several extensions have been investigated: Mechanisms for measuring trust and reputation [96], advanced attacks on such trust communities and corresponding countermeasures [97], accusation-based strategies to identify misbehaving subsystems and to establish forgiveness solutions [98], and robust self-monitoring mechanisms at run-time [99]. Furthermore, [100] demonstrated that computational trust and forgiveness techniques can result in improved reliability and reduced overhead.…”

Clonal plasticity

“…Most multi-agent system based adaptation relies on the distributed and autonomous nature of the constituent agents [11], but there is little literature on how to configure the agents themselves. Applications such as collision avoidance [6], grid management [16] use collaborative learning, and hierarchically deferring to a controller agent, to achieve the application's objectives. The most popular technique of self-configuration in multi-agent systems is reinforcement-learning [4].…”