A Memory Configuration Method for Virtual Machine Based on User Preference in Distributed Cloud

Weijia Jia, Shukun Liu, Weijia Jia 0001, Xianmin PanKsii Trans. Internet Inf. Syst 2018 · DOI

Abstract

It is well-known that virtualization technology can bring many benefits not only to users but also to service providers. From the view of system security and resource utility, higher resource sharing degree and higher system reliability can be obtained by the introduction of virtualization technology in distributed cloud. The small size time-sharing multiplexing technology which is based on virtual machine in distributed cloud platform can enhance the resource utilization effectively by server consolidation. In this paper, the concept of memory block and user satisfaction is redefined combined with user requirements. According to the unbalanced memory resource states and user preference requirements in multi-virtual machine environments, a model of proper memory resource allocation is proposed combined with memory block and user satisfaction, and at the same time a memory optimization allocation algorithm is proposed which is based on virtual memory block, makespan and user satisfaction under the premise of an orderly physical nodes states also. In the algorithm, a memory optimal problem can be transformed into a resource workload balance problem. All the virtual machine tasks are simulated in Cloudsim platform. And the experimental results show that the problem of virtual machine memory resource allocation can be solved flexibly and efficiently.

Introduction

Virtualization technology is a cornerstone of modern cloud computing, offering significant benefits in resource utility and system reliability. By allowing multiple virtual machines (VMs) to share physical hardware, service providers can achieve higher resource consolidation. However, while CPU and I/O scheduling have seen extensive optimization, memory allocation remains a critical bottleneck, particularly in distributed cloud environments.

In a distributed cloud, tasks are spread across geographically dispersed servers. Traditional memory management often focuses on single-machine optimization and fails to account for "user preference"—the trade-off between service cost and execution speed—and varying task priorities. This oversight leads to unbalanced resource distribution, where some nodes are over-utilized while others remain idle, ultimately degrading user satisfaction.

System Model and Memory Abstraction

The proposed approach begins by redefining the relationship between physical memory and virtual instances. Instead of viewing memory as a fragmented resource tied to specific hardware, the authors propose a "Whole logic view of all physical memory."

Logical view of physical memory abstraction in a distributed cloud environment.

As shown in Figure 1, various physical devices (servers, laptops, mobile devices) contribute to a unified logical memory pool. This pool is then partitioned into virtual memory blocks that serve a series of virtual machines ($VM_1$ to $VM_n$). This abstraction allows the system to treat memory allocation as a workload balancing problem across a set of "pseudo-processors."

Methodology: The ISAPS Algorithm

The core of this research is the Improved Satisfaction and Priority Scheduling (ISAPS) algorithm. This heuristic approach aims to optimize memory allocation based on three primary pillars:

  1. User Preference: A mathematical model that balances the time consumed by a task against the financial cost to the user.
  2. Memory Block Usability: Defined as the probability ($exp(-\lambda t)$) that a system can provide the necessary space for a task, accounting for the inherent instability of distributed nodes.
  3. Task Priority: Ensuring that critical or high-paying tasks receive resources preferentially.

To formalize these concepts, the authors utilize several signals to calculate expected running times and finishing times.

Definitions of signals and variables used in the ISAPS mathematical model.

As detailed in Figure 2, variables such as $ZET_{ij}$ (expected running time) and $ZZT_j$ (early using time of a node) are used to calculate the finishing time ($FT_{ij}$) of a task on a specific VM. The ISAPS algorithm introduces a weight parameter, $\psi$ (psi), which serves as a tuning knob for the system. When $\psi = 1$, the system prioritizes user satisfaction (speed); when $\psi = 0$, it prioritizes VM priority (cost/importance).

Experimental Evaluation

The researchers evaluated the ISAPS algorithm using the CloudSim 3.0 platform. To ensure realistic results, they utilized trace data from the Lawrence Livermore National Laboratory (LLNL), specifically logs from a large Linux cluster. These logs were converted into eight types of VM requests and eight types of logical physical machines.

Key Results

The ISAPS algorithm was compared against four standard scheduling methods: Random, Round-Robin, List Scheduling, and First-Come-First-Served (FCFS).

  • Imbalance Degree: ISAPS consistently achieved the lowest imbalance degree across various task durations. This indicates that the algorithm effectively prevents resource hotspots and ensures a more uniform distribution of memory load.
  • Makespan Efficiency: The "makespan" (total time to complete all tasks) was significantly reduced. For long-duration tasks (DT > 80), ISAPS outperformed the Random algorithm by nearly 15%.
  • Workload Capacity: ISAPS demonstrated a more stable "capacity makespan" curve, suggesting it handles increasing workloads more gracefully than traditional round-robin or list scheduling approaches.

Conclusion and Future Work

This paper presents a flexible memory configuration model that bridges the gap between technical resource management and user-centric service requirements. By treating memory as a logical pool and applying the ISAPS algorithm, distributed clouds can achieve better load balancing and higher user satisfaction.

However, limitations remain. The current model primarily focuses on memory, whereas a holistic optimizer should simultaneously consider CPU and I/O constraints. Additionally, while the LLNL data provides a strong foundation, future research should test the algorithm in live production environments to account for real-world network latency and "noisy neighbor" interference. Further refinements could also involve using machine learning to dynamically adjust the $\psi$ parameter in response to real-time traffic fluctuations.

Summary updated

BibTeX
@article{jia2018memory,
  author = {Weijia Jia and Shukun Liu and Weijia Jia and Xianmin Pan},
  title = {{A Memory Configuration Method for Virtual Machine Based on User Preference in Distributed Cloud}},
  journal = {Ksii Trans. Internet Inf. Syst},
  year = {2018},
  doi = {10.3837/tiis.2018.11.004},
}
Figure