Abstract
Wireless network virtualization (WNV) provides a novel paradigm shift in the fifth-generation (5G) system, enabling more efficient utilization of network resources. This paper investigates the minimization of network overheads while satisfying Quality of Service (QoS) requirements regarding overflow probability by jointly considering cache space and time-frequency resource allocation. We propose an online adaptive virtual resource allocation algorithm with multiple time-scales based on the Auto-Regressive Moving Average (ARMA) prediction method. This approach addresses the uncertainties of traffic and information feedback delays inherent in traditional methods. Simulation results demonstrate significant gains in reducing bit loss rates and improving physical resource utilization.
Introduction
In the era of 5G, Wireless Network Virtualization (WNV) allows multiple service providers to share the same physical infrastructure. The primary challenge for infrastructure providers is "slicing" resources—specifically cache space and time-frequency bandwidth—efficiently.
Traditional allocation methods often fall into two traps: over-provisioning, which wastes resources and lowers utilization, or under-provisioning, which fails to handle traffic spikes and leads to queue overflows and data loss. Furthermore, traffic is inherently volatile, and information feedback delays often lead to "irrational" allocation decisions when cache and bandwidth are treated as independent silos. This paper introduces RALF (Resource Allocation based on Load Forecasting), an online adaptive algorithm designed to optimize these coupled resources.
System Architecture
The proposed framework utilizes a centralized management platform to coordinate resource distribution between wireless virtual network users and the physical resource pool.
As illustrated in the architecture, the system consists of three main components:
- Wireless Virtual Network Users: Diverse service requests from various virtual networks (e.g., mobile devices, IoT sensors).
- Virtual Network Management Platform: This core entity performs load analysis (using ARMA for prediction), manages long-term cache reservation, and schedules short-term bandwidth allocation.
- Physical Resource Pool: The shared pool consisting of time-frequency Resource Blocks (RBs) and physical cache space.
Methodology: The RALF Algorithm
The RALF algorithm employs a multi-time-scale resource scheduling mechanism that treats cache and bandwidth as complementary resources:
- Long Time-Scale (Cache Reservation): The system uses an Auto-Regressive Moving Average (ARMA) model to predict future load trends based on historical data. This allows the network to proactively reserve cache portions, ensuring sufficient buffer for predicted surges without over-committing physical memory.
- Short Time-Scale (Bandwidth Scheduling): Within each long-term period, the system performs rapid scheduling of time-frequency RBs. Because exact traffic distribution is often unknown, the Large Deviation Principle (LDP) is applied to estimate the probability of a queue overflow.
- Prioritization: Virtual networks are prioritized based on their "overflow residual time," targeting the networks at the highest risk of failing their QoS targets first.
Performance Evaluation
The study relies on simulated traffic data to evaluate the algorithm, modeling the arrival process as a Poisson distribution with a bit arrival rate of 58.7 kbit/ms. RALF was compared against four traditional schemes: Static Conservative Cache Allocation (SCCA), Static Abundant Cache Allocation (SACA), Hard Slicing (HS), and Slice with Fixed Prioritization (SFP).
Key results include:
- Reduced Costs: RALF significantly lowered resource leasing overhead compared to "abundant" allocation schemes (SACA), as it only reserves what is predicted to be necessary.
- Lower Bit Loss: Despite using fewer resources, RALF achieved a much lower bit loss rate than "conservative" schemes (SCCA) by dynamically expanding capacity during predicted peaks.
- High Utilization: Simulation results showed that RALF maintained physical resource utilization (often above 80%) while keeping bit loss at an ideal level, outperforming both Hard Slicing and Fixed Prioritization.
- Scalability: As the number of virtual networks increased, RALF's cost grew much more slowly than traditional fixed-allocation methods.
Conclusion and Future Work
RALF provides a robust, proactive approach to 5G resource management by coupling cache and bandwidth allocation through multi-time-scale prediction and risk assessment. By using the Large Deviation Principle, the authors provide a way to estimate overflow probabilities even when underlying traffic distribution is complex.
Future work will focus on refining the cache reservation policy for more diverse 5G application scenarios (e.g., ultra-low latency vs. high-throughput video) and validating the framework against real-world 5G traffic traces to confirm its effectiveness against non-Poissonian "bursty" traffic.
Summary updated