A Trust-Based Agent Learning Model for Service Composition in Mobile Cloud Computing Environments

Wenjuan Li, Jian Cao, Keyong Hu, Jie Xu, Rajkumar BuyyaIEEE Access 2019 · OpenReview

Abstract

Mobile Cloud Computing (MCC) faces significant challenges due to resource constraints, open environments, and inherent uncertainty, leading to security risks and unpredictable Quality of Service (QoS). This paper proposes a Trust-based Agent Learning Model for Service Composition (TALMSC) to address these issues. The model utilizes a three-layered multi-agent architecture, a fuzzy comprehensive evaluation trust model, and an improved fuzzy clustering learning algorithm. Experimental results demonstrate that TALMSC enhances transaction success rates and user satisfaction, even in environments with high proportions of malicious providers.

Introduction

The proliferation of mobile devices has driven the growth of Mobile Cloud Computing (MCC), which allows resource-constrained devices to offload tasks to cloud-based infrastructures. However, the dynamic and open nature of MCC introduces volatility. Users often struggle to find reliable services that match their subjective preferences, while providers may fail to deliver promised QoS or act maliciously. Traditional service composition methods often fall short in these autonomous, collaborative environments.

The TALMSC Architecture

To navigate these challenges, the researchers developed the Trust-based Agent Learning Model for Service Composition (TALMSC). The model is built upon a multi-agent system (MAS) framework, which is well-suited for simulating the autonomous interactions of cloud entities.

Figure 5: The three-layered multi-agent architecture of the TALMSC model, showing the interaction between User, Broker, and Provider Agents.

As shown in Figure 5, the architecture consists of three primary agent types:

  1. User Agents: Represent cloud customers, managing service requests and making final trust decisions based on their specific needs.
  2. Provider Agents: Represent cloud resource providers, managing the underlying resource pool and offering services to the market.
  3. Broker Agents: Act as intelligent intermediaries. They are equipped with learning modules to analyze user preferences and trust management modules to filter providers.

Methodology: Trust and Learning

The core of the TALMSC model lies in its ability to handle "fuzzy" information—data that is not strictly binary but exists on a spectrum of quality or preference.

Fuzzy Comprehensive Evaluation (FCE) Trust Model

Trust is not a monolithic score. A provider might be excellent for storage but poor for high-speed computation. The FCE model evaluates providers across multiple dimensions (CPU, RAM, Storage, Bandwidth, Price). It calculates an integrated trust score by combining Direct Trust (derived from the broker's own historical interactions) and Recommendation Trust (aggregated feedback from other agents). By using fuzzy membership degrees, the model categorizes providers into levels such as "highly trusted," "normal," or "untrusted."

Two-Stage Improved Fuzzy C-Means (FCM) Learning

User requirements are often vague (e.g., "I want a fast service"). To bridge the gap between vague requests and specific service attributes, Broker Agents employ a two-stage learning process:

  1. Improved Maxmin Algorithm: Standard clustering algorithms like FCM are sensitive to initial conditions. The TALMSC model uses a Maxmin algorithm to intelligently select initial cluster centers based on data density, avoiding "noise" and local optima.
  2. FCM Clustering: The broker then applies FCM to refine these clusters, learning the latent preferences of users from their transaction history. This allows the broker to proactively suggest services that align with a user's subjective priorities.

Evaluation and Results

The model was evaluated using NetLogo for large-scale ecosystem simulation and the JADE (Java Agent Development Framework) for multi-agent implementation. Service parameters were grounded in real-world data from Ali-Cloud’s Elastic Compute Service (ECS).

Key findings include:

  • Resilience: The model maintained high transaction success rates even when 95% of the providers in the simulation were malicious.
  • Efficiency: The Improved-FCM algorithm converged to stable user preference profiles significantly faster than standard K-Means or basic FCM.
  • Satisfaction: User satisfaction scores were consistently higher in the TALMSC model compared to non-learning or standard clustering approaches, proving the value of the broker's learning module.

Conclusion

The TALMSC model provides a robust framework for service composition in the unpredictable environment of mobile cloud computing. By integrating fuzzy trust management with autonomous agent learning, it effectively mitigates security risks and improves the alignment between cloud services and user expectations. Future research will focus on integrating predictive forecasting and exploring market specialization among brokers.

Summary updated

Abstract

Mobile cloud computing has the features of resource constraints, openness, and uncertainty which leads to the high uncertainty on its quality of service (QoS) provision and serious security risks. Therefore, when faced with complex service requirements, an efficient and reliable service composition approach is extremely important. In addition, preference learning is also a key factor to improve user experiences. In order to address them, this paper introduces a three-layered trust-enabled service composition model for the mobile cloud computing systems. Based on the fuzzy comprehensive evaluation method, we design a novel and integrated trust management model. Service brokers are equipped with a learning module enabling them to better analyze customers’ service preferences, especially in cases when the details of a service request are not totally disclosed. Because traditional methods cannot totally reflect the autonomous collaboration between the mobile cloud entities, a prototype system based on the multi-agent platform JADE is implemented to evaluate the efficiency of the proposed strategies. The experimental results show that our approach improves the transaction success rate and user satisfaction.
BibTeX
@inproceedings{li2019trust,
  author = {Wenjuan Li and Jian Cao and Keyong Hu and Jie Xu and Rajkumar Buyya},
  title = {{A Trust-Based Agent Learning Model for Service Composition in Mobile Cloud Computing Environments}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=zI4ijCBl8r},
}
Figure