Research on the Optimal Resource Allocation Policy based on User Characteristics and Behavior under the Environment of Cloud

Rong Wang, Rao Bin2015 · DOI

Abstract

In this paper, we research on the optimal resource allocation policy based on user characteristics and behavior under the environment of cloud. Computing resources and the services are in accordance with the rapid demand to provide to the user to realize through the Internet anytime and calculation mode of easy access to a shared resource pool. The main content of cloud computing for the management of computing is power, storage, platform and services. How to finalize optimal resource allocation policy has significant meaning. Our research deal with this issue well, the numerical simulation illustrate the feasibility of the proposed method.

Introduction

As cloud computing scales to a massive magnitude, traditional resource allocation algorithms—originally designed for static grid computing—are increasingly failing to keep up with modern demands. The core challenge lies in the high volatility of user demand. In many contemporary cloud environments, resources are over-provisioned to ensure high availability, which inadvertently leads to low server utilization and massive energy waste.

This research identifies a specific gap in current methodologies: most systems do not account for the "human element," specifically the predictable habits and individual characteristics of users. Without understanding how and when users access services, a system cannot effectively balance the trade-off between performance and power consumption. This paper proposes a dynamic allocation strategy that leverages virtualization and user behavior prediction to maximize resource efficiency.

Methodology

The proposed system utilizes a multi-layered architecture to optimize the cloud environment across Infrastructure as a Service (IaaS), Platform as a Service (PaaS), and Software as a Service (SaaS) layers.

The multi-layered cloud service architecture showing the interaction between customers and the SaaS, PaaS, and IaaS layers.

User Behavior and Collaborative Filtering

The system analyzes individual user habits against group "threshold values" to predict future resource needs. Borrowing from recommendation system theory, a collaborative filtering algorithm is employed to predict user needs based on "neighbor sets"—groups of users with similar behavior patterns. By modeling state changes using transition probabilities, the system can efficiently predict the next phase of resource demand if the next state depends primarily on the current state.

Access "Heat" and Node Management

For storage optimization, data files are divided into fixed-size blocks. The system calculates the "heat" (frequency of access) for these blocks. "Cold" nodes—those hosting rarely accessed data—are transitioned into dormancy or closed to save energy. This approach addresses the energy-availability paradox: while high availability usually requires redundant active nodes, using access heat allows the system to determine which redundant nodes are actually necessary in real-time.

Virtualization Decoupling

By utilizing Virtual Machines (VMs), the execution environment is isolated from the physical hardware. This allows the system to package and transmit execution environments to different physical nodes dynamically. This decoupling is the primary mechanism for path optimization, allowing workloads to be moved away from nodes slated for power-down without interrupting the application.

Results and Discussion

The feasibility of the behavior-based policy was validated through numerical simulations involving 21 different subjects or scenarios.

Statistical analysis of scores across 21 subjects, demonstrating the system's ability to maintain service quality while adjusting resource distribution.

Key Findings

  • Feasibility: The simulation results, illustrated via the box plots in Figure 3, indicate that the system can maintain service quality while dynamically adjusting resource distribution based on user characteristics.
  • Energy Efficiency: By converting access features into "file access heat," the system effectively shut down unnecessary nodes. This resulted in a significant reduction in energy waste without affecting the user's perceived quality of service.
  • Utilization Rates: The strategy improved the overall utilization rate of hardware resources by ensuring that the contribution rate of customer satisfaction was maximized relative to the hardware power being drawn.

Conclusion and Future Work

This research demonstrates that user behavior is not random noise but a quantifiable predictor that allows cloud environments to be proactive rather than reactive. However, several challenges remain. The "cold start" problem persists, where the algorithm struggles with new users who lack historical data. Additionally, data sparsity in the user-project rating matrix can lead to inaccurate similarity measures.

Future work will focus on refining the transition probability models and developing more robust methods to handle sparse data in the user characteristic matrix, moving toward a more systematic framework for cloud resource allocation.

Summary updated

Abstract

In this paper, we research on the optimal resource allocation policy based on user characteristics and behavior under the environment of cloud. Computing resources and the services are in accordance with the rapid demand to provide to the user to realize through the Internet anytime and calculation mode of easy access to a shared resource pool. The main content of cloud computing for the management of computing is power, storage, platform and services. How to finalize optimal resource allocation policy has significant meaning. Our research deal with this issue well, the numerical simulation illustrate the feasibility of the proposed method.
BibTeX
@misc{wang2015research,
  author = {Rong Wang and Rao Bin},
  title = {{Research on the Optimal Resource Allocation Policy based on User Characteristics and Behavior under the Environment of Cloud}},
  year = {2015},
  doi = {10.2991/iemb-15.2015.32},
}
Figure