[Paper Review] A Resource Management Protocol for Mobile Cloud Using Auto-Scaling

Chathura M. Sarathchandra Magurawalage, Kun Yang+3 more|arXiv (Cornell University)|Jan 2, 2017

IoT and Edge/Fog Computing26 references3 citations

TL;DR

This paper proposes a centralized resource management protocol for Mobile Cloud Computing (MCC) and Cloud Radio Access Networks (C-RAN) that enables dynamic task offloading and auto-scaling of virtual machine (VM) resources. It reveals that vertical auto-scaling—especially for CPU and storage—incurs significant delays (over 50 seconds in some cases), making real-time scaling impractical for delay-sensitive applications.

ABSTRACT

Cloud radio access networks (C-RAN) and Mobile Cloud Computing (MCC) have emerged as promising candidates for the next generation access network techniques. MCC enables resource limited mobile devices to offload computationally intensive tasks to the cloud, while C-RAN offers a technology that addresses the increasing mobile traffic. In this paper, we propose a protocol for task offloading and for managing resources in both C-RAN and mobile cloud together using a centralised controller. Experiments on resource management using cloud auto-scaling shows that resource (CPU, RAM, Storage) scaling times vary.

Motivation & Objective

• Address the challenge of efficiently managing computing and communication resources in hybrid Mobile Cloud and C-RAN architectures.
• Design a unified protocol for task offloading and resource allocation across mobile cloud and C-RAN using a centralized controller.
• Investigate the performance of vertical auto-scaling for CPU, RAM, and storage in mobile cloud environments.
• Assess the impact of auto-scaling delay on real-time application viability in mobile cloud deployments.

Proposed method

• Proposes a centralized controller to manage both mobile cloud VMs and C-RAN baseband units (BBUs) via a unified signaling protocol.
• Employs a simple, standardized packet header for both task offloading and resource management signaling.
• Conducts empirical evaluation of VM resize performance (upscaling/downscaling) for CPU, RAM, and storage under continuous and non-continuous scaling scenarios.
• Uses KVM-based virtualization in a testbed to measure actual scaling delays across different VM disk sizes and resource types.
• Applies second-order polynomial regression to model scaling delay trends based on base VM disk size and resource type.
• Analyzes the influence of VM size, scaling method (continuous vs. non-continuous), and cloud environment on auto-scaling latency.

Experimental results

Research questions

• RQ1 How does the auto-scaling delay vary across different VM resources (CPU, RAM, storage) in mobile cloud environments?
• RQ2 What is the impact of scaling method (continuous vs. non-continuous) on VM resize time for different resource types?
• RQ3 How does the base VM disk size affect the time delay during VM resizing operations?
• RQ4 To what extent does auto-scaling delay hinder real-time applicability for latency-sensitive mobile applications?
• RQ5 Can performance trends in VM resizing be modeled to inform smarter auto-scaling decision-making in mobile cloud systems?

Key findings

• Auto-scaling delays for CPU and storage resources exceed 50 seconds in some cases, making real-time scaling impractical for delay-sensitive applications.
• Scaling delay increases with VM disk size in continuous scaling scenarios, particularly for storage and CPU resources.
• No clear trend was observed for RAM resizing delays due to limited sample size and hardware constraints (maximum 18GB memory), though regression models suggest decreasing trends.
• The scaling delay depends not only on the resource type but also on the scaling method (upscaling/downscaling) and VM configuration.
• VM resize performance varies significantly based on underlying virtualization technology and cloud environment, necessitating platform-specific tuning.
• Future auto-scaling algorithms must consider both VM-specific characteristics (e.g., disk size) and environmental factors (e.g., total VM count) for effective decision-making.