A Framework to Quantify the Benefits of Network Functions Virtualization in Cellular Networks

Zafar Ayyub Qazi, Vyas Sekar, Samir Ranjan DasarXiv 2014 · arXiv

TL;DR

This paper introduces a framework to help cellular network operators systematically evaluate the cost-benefit trade-offs of different Network Functions Virtualization (NFV) deployment strategies. By modeling provisioning costs, traffic patterns, and performance constraints, it provides a quantitative approach to optimize NFV adoption in complex cellular networks.

What Problem Does This Paper Address?

Cellular networks today face significant challenges due to high capital and operational expenses associated with inflexible, hardware-based network functions. Network Functions Virtualization (NFV) promises to alleviate these issues by virtualizing these functions onto commodity hardware, akin to cloud computing. However, the "design space" for NFV is vast, encompassing choices in platform types (e.g., cloud, flexible hardware), provisioning strategies (where to place virtualized functions), and traffic distribution. Operators need a systematic way to navigate this complexity and quantify the potential benefits of various NFV instantiations before making costly infrastructure decisions. This paper aims to fill that gap by providing a framework for such an evaluation.

Methods

The authors propose a formal optimization framework to model the cost of provisioning network functions to meet time-varying processing and performance requirements. The core of their approach involves defining a set of "control variables" that capture the key decisions an operator must make:

  • Platform Type: Whether to use dedicated hardware ("Single"), flexible commodity hardware ("FlexHW"), or a public cloud.
  • Provisioning and Placement: Deciding if and where to deploy NF instances, and how much resource (e.g., CPU, memory) to provision. This includes dynamic provisioning for elastic options like Cloud and FlexHW.
  • Load Distribution: How to route traffic across different NF instances to meet service chain requirements and performance constraints.

The framework formulates an objective function to minimize total provisioning cost, which includes fixed costs for platform deployment, hardware costs for provisioned resources, and dynamic costs for epoch-specific resource usage. This is then constrained by:

  • Resource Provisioning: Ensuring that the provisioned resources on each platform are sufficient to handle the total load from all NF functions it hosts.
  • Traffic Distribution: Modeling flow conservation to ensure traffic is routed correctly through the defined service chains and that all traffic is accounted for.
  • Performance Bounds: Incorporating constraints on metrics like end-to-end latency for specific traffic classes.

The problem is cast as an Integer Linear Program (ILP) and solved using CPLEX.

Datasets

The paper uses a combination of:

  • Internet2/Abilene topology: A standard network topology for evaluating network performance.
  • Gravity model: Used to generate baseline traffic demand, simulating city populations and randomized variability.
  • Hypothetical cost numbers: Due to the lack of publicly available, precise cost data for NFV platforms, the authors derive ballpark figures based on industry reports and common assumptions for cloud services (e.g., Amazon EC2), commodity servers, and specialized devices.

Key Insights and Ideas

  • NFV Design Space is Complex: The paper effectively highlights that NFV is not a monolithic solution but a spectrum of choices across platform, provisioning, and distribution.
  • Cost-Benefit Trade-offs are Crucial: The framework demonstrates that the "optimal" NFV strategy is highly dependent on factors like traffic volume, variability, and performance requirements (especially latency).
  • Hybrid Models Offer Flexibility: The study suggests that hybrid deployment models, combining different platform types, can often provide a good balance of cost and performance, especially under varying traffic conditions.
  • Latency is a Key Differentiator: Performance constraints, particularly latency, can significantly influence deployment decisions, sometimes forcing the use of more expensive but lower-latency solutions like FlexHW or even dedicated hardware, even if cloud solutions are cheaper.
  • Quantification is Essential: The paper's core contribution is providing a quantitative method to move beyond qualitative discussions about NFV benefits and enable data-driven decision-making for operators.

Key Results

The paper presents illustrative use cases and sensitivity analyses:

  • Provisioning Cost Optimization: In a scenario with high traffic volume and low fixed costs for cloud, a cloud-only solution was found to be the most cost-effective.
  • Performance Constraints Drive Hybrid Solutions: When a latency constraint was introduced for voice traffic, a hybrid solution (partially on fixed hardware, partially on cloud) became the most cost-effective, avoiding higher costs of pure cloud and the limitations of pure fixed hardware.
  • FlexHW's Role: The sensitivity analysis showed that using flexible hardware can minimize deployment costs in many scenarios.
  • Impact of Cloud Cost: As cloud costs decrease, cloud-only strategies become more viable, potentially leading to an optimal strategy using only public cloud resources.
  • Impact of Setup/OPEX: Increasing setup and operational costs for fixed infrastructure incentivizes less pre-provisioning and potentially more cloud adoption.
  • Performance Gap: A larger performance gap between virtualized appliances and specialized hardware makes hybrid models more attractive.

Limitations and Future Work

Limitations:

  • Simplified Cost Models: The paper acknowledges the use of ballpark figures for costs due to the lack of precise, publicly available data.
  • Basic Performance Constraints: The initial framework focuses primarily on end-to-end latency as the performance constraint, while real-world networks have more complex requirements.
  • Simplified Traffic Models: The traffic models used are based on gravity models and randomized variability, which may not fully capture the nuances of real cellular traffic patterns.
  • Limited Scope of "Hybrid": In one example, the hybrid model was restricted to FlexHW and Single platforms, not encompassing all three types.

Future Work:

  • Operator Collaboration: Actively working with operators to understand new deployment opportunities and obtain more realistic datasets and policy requirements.
  • Advanced Policy Considerations: Incorporating other policy considerations, such as co-locating roaming users with other users on the same hardware.
  • Exploring Dynamicity: Further investigating how the dynamicity and flexibility of NFV can enable new deployment and service opportunities.
  • More Complex Traffic and Performance Models: Expanding the framework to handle more sophisticated traffic patterns and a wider range of performance metrics.

Summary updated

BibTeX
@misc{qazi2014framework,
  author = {Zafar Ayyub Qazi and Vyas Sekar and Samir Ranjan Das},
  title = {{A Framework to Quantify the Benefits of Network Functions Virtualization in Cellular Networks}},
  year = {2014},
  eprint = {1406.5634},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1406.5634},
}
Figure