Stochastic Game Analysis and Latency Awareness for Self-Adaptation
Abstract
Although different approaches to decision-making in self-adaptive systems have shown their effectiveness in the past by factoring in predictions about the system and its environment (e.g., resource availability), no proposal considers the latency associated with the execution of tactics upon the target system. However, different adaptation tactics can take different amounts of time until their effects can be observed. In reactive adaptation, ignoring adaptation tactic latency can lead to suboptimal adaptation decisions (e.g., activating a server that takes more time to boot than the transient spike in traffic that triggered its activation). In proactive adaptation, taking adaptation latency into account is necessary to get the system into the desired state to deal with an upcoming situation. In this paper, we introduce a formal analysis technique based on model checking of stochastic multiplayer games (SMGs) that enables us to quantify the potential benefits of employing different types of algorithms for self-adaptation. In particular, we apply this technique to show the potential benefit of: (i) considering adaptation tactic latency in proactive adaptation algorithms, and (ii) making available additional tactics in the repertoire employed to adapt a system. Our results show that factoring in tactic latency in decision making, not only improves the outcome of adaptation, but also enables algorithms to fully exploit the set of available tactics for adaptation. We also present an algorithm to do proactive adaptation that considers tactic latency, and show that it achieves higher utility than an algorithm that under the assumption of no latency is optimal.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2014
- Accession Number
- ADA611042
Entities
People
- Bradley Schmerl
- David Garlan
- Gabriel A. Moreno
- Javier Cámara
Organizations
- Carnegie Mellon University