Towards all-IP wireless networks: architectures and resource management mechanism

Majid Ghaderi, Raouf BoutabaInt. J. Wirel. Mob. Comput 2007 · DOI

TL;DR

This paper proposes a unified architecture for integrating diverse wireless networks (like 3G and Wi-Fi) into a single all-IP network. It introduces a novel prediction-based admission control mechanism that dynamically manages network resources to ensure quality of service for mobile users, even with unpredictable traffic patterns.

What Problem Does This Paper Address?

The proliferation of wireless technologies (3G, Wi-Fi, satellite) has created a fragmented landscape for mobile users. Seamlessly transitioning between these networks while maintaining a consistent quality of service (QoS) is a significant challenge. This heterogeneity arises from differences in access technologies, network architectures, protocols, and user service demands. Existing IP networks primarily offer "best-effort" service, which is insufficient for the diverse QoS requirements of modern applications. Furthermore, traditional resource management techniques struggle with the dynamic and often unpredictable nature of wireless environments, leading to inefficient resource utilization and poor user experience (e.g., dropped connections). This paper aims to bridge this gap by proposing architectures and a resource management mechanism for a unified all-IP wireless network.

Methods

The paper proposes three architectures for all-IP wireless networks:

  • IntServ-Based Architecture (ISB): Combines DiffServ in the core network with IntServ in the wireless access network. This is suitable for low-bandwidth 3G cellular networks with strong resource management capabilities.
  • DiffServ-Based Architecture (DSB): Uses DiffServ in both the core and wireless access networks. This is targeted for high-bandwidth wireless LANs with less stringent resource management needs.
  • Integrated All-IP Architecture (AIP): A hybrid approach that combines the strengths of ISB and DSB to facilitate seamless integration of different wireless technologies like 3G and Wi-Fi.

For resource management within the AIP architecture, the paper introduces a hierarchical admission control mechanism called Prediction-based Admission Control (PrBAC). Key aspects of PrBAC include:

  • Measurement-based: It relies on online measurements rather than predefined traffic models.
  • MMSE Predictor: Uses a Minimum Mean Square Error (MMSE) predictor to forecast bandwidth requirements for incoming handoffs and new connections. This predictor is on-line and does not require specific traffic modeling, making it suitable for IP networks with non-Poisson traffic.
  • Hierarchical Control: Operates at two levels: base stations (BS) for intra-domain handoffs and gateways (GW) for inter-domain handoffs.
  • Dynamic Reservation: Reserves bandwidth for predicted handoffs, dynamically adjusting thresholds based on measured traffic.
  • Priority Handling: Prioritizes higher-priority traffic (e.g., expedited forwarding, assured forwarding) over best-effort traffic when necessary.

Datasets

The paper uses a real Ethernet traffic trace from Bellcore (collected by Leland et al.) for evaluating the accuracy of the MMSE predictor against other self-similar traffic models (fGn, FARIMA, GARMA). For evaluating the performance of the PrBAC scheme, simulations were conducted using two types of generated traffic:

  1. Poisson Traffic: Inter-arrival times for new and handoff connections are exponentially distributed.
  2. Non-Poisson Traffic: Inter-arrival times are generated using an autoregressive model of order one (AR(1)).

Key Insights and Ideas

  • Unified All-IP is Achievable: The paper demonstrates that integrating diverse wireless networks into a single IP-based framework is feasible through well-designed architectures like AIP.
  • Dynamic Prediction is Crucial: Traditional static admission control methods struggle with the inherent variability of wireless traffic. PrBAC's strength lies in its ability to dynamically predict and adapt to changing network conditions.
  • MMSE Predictor's Versatility: The MMSE predictor's effectiveness in forecasting bandwidth needs without relying on specific traffic models is a significant advantage for IP networks, which often exhibit complex traffic patterns.
  • Hierarchical Control Balances Efficiency and Responsiveness: The two-level admission control (BS and GW) allows for efficient local management while also handling inter-domain mobility, reducing signaling overhead and enabling faster handoffs.
  • Trade-off Between Blocking and Dropping: The core challenge in admission control is balancing connection blocking probability (rejecting new connections) with connection dropping probability (dropping active connections). PrBAC aims to minimize dropping while keeping blocking low.

Key Results

  • Predictor Accuracy: The MMSE predictor's accuracy (measured by SNR⁻¹) was found to be reasonably close to that of the best predictor (FARIMA) in experiments with a real Ethernet traffic trace.
  • PrBAC Performance vs. CDAC:
    • Poisson Traffic: PrBAC consistently achieved higher call completion probabilities than the static Cell-Domain Admission Control (CDAC) scheme, especially under increasing load. Both schemes could provide a guaranteed dropping probability.
  • Non-Poisson Traffic: PrBAC also outperformed CDAC in terms of connection completion probability. While PrBAC achieved a significantly lower dropping probability than CDAC, the final connection completion probabilities were similar, suggesting that connection blocking was the dominant factor.
  • Dynamic Nature Advantage: The dynamic nature of PrBAC, which tunes its admission threshold every control period, leads to better bandwidth utilization compared to the static CDAC.

Limitations and Future Work

  • Real Traffic Complexity: The authors acknowledge that real traffic patterns are more complex than the AR(1) model used. Further research is needed to ensure PrBAC performs optimally with highly complex traffic.
  • Reactive PrBAC: While PrBAC aims to be reactive, the paper notes that more investigation is required to make it more reactive to discrepancies in predicted vs. actual dropping probabilities to achieve even better performance.
  • Adaptive Applications: The paper highlights the need for adaptive applications that can adjust their bandwidth requirements based on network capabilities and service negotiation protocols for seamless roaming.
  • Vertical Handoff: The paper addressed hard vertical handoffs. Future work could explore soft handoff schemes for overlapping coverage areas and sophisticated decision-making for network selection.
  • Other Issues: Security, pricing, and billing are identified as important areas for future consideration in the complete realization of the AIP architecture.

Summary updated

BibTeX
@article{ghaderi2007towards,
  author = {Majid Ghaderi and Raouf Boutaba},
  title = {{Towards all-IP wireless networks: architectures and resource management mechanism}},
  journal = {Int. J. Wirel. Mob. Comput.},
  year = {2007},
  doi = {10.1504/ijwmc.2007.016720},
}
Figure