[Paper Review] Multi-Timescale Online Optimization of Network Function Virtualization for Service Chaining

Xiaojing Chen, Wei Ni+5 more|arXiv (Cornell University)|Apr 19, 2018

Software-Defined Networks and 5G21 references3 citations

TL;DR

This paper proposes a multi-timescale, fully decentralized online optimization framework for Network Function Virtualization (NFV) service chaining, using a stochastic dual gradient method to decouple VM decisions and minimize time-average cost while stabilizing service backlogs. It achieves a cost-backlog tradeoff of $[ heta, heta^{-1}]$ and improves to $[ heta, heta^{-1} ext{log}^2(1/ heta)]$ via a learn-and-adapt strategy, reducing NFV cost by 30% and delay by 83% compared to benchmarks.

ABSTRACT

Network Function Virtualization (NFV) can cost-efficiently provide network services by running different virtual network functions (VNFs) at different virtual machines (VMs) in a correct order. This can result in strong couplings between the decisions of the VMs on the placement and operations of VNFs. This paper presents a new fully decentralized online approach for optimal placement and operations of VNFs. Building on a new stochastic dual gradient method, our approach decouples the real-time decisions of VMs, asymptotically minimizes the time-average cost of NFV, and stabilizes the backlogs of network services with a cost-backlog tradeoff of $[ε,1/ε]$, for any $ε> 0$. Our approach can be relaxed into multiple timescales to have VNFs (re)placed at a larger timescale and hence alleviate service interruptions. While proved to preserve the asymptotic optimality, the larger timescale can slow down the optimal placement of VNFs. A learn-and-adapt strategy is further designed to speed the placement up with an improved tradeoff $[ε,\log^2(ε)/{\sqrtε}]$. Numerical results show that the proposed method is able to reduce the time-average cost of NFV by 30\% and reduce the queue length (or delay) by 83\%, as compared to existing benchmarks.

Motivation & Objective

• Address the challenge of coupled, real-time decisions in NFV service chaining where virtual network functions (VNFs) must be placed across VMs in a specific order.
• Overcome limitations of centralized or heuristic approaches by enabling fully decentralized, online decision-making under stochastic service arrivals and dynamic pricing.
• Stabilize network service backlogs while minimizing time-average operational cost in large-scale, dynamic NFV environments.
• Introduce a multi-timescale design to reduce service interruptions from frequent VNF re-placements, balancing optimality and stability.
• Develop a learn-and-adapt strategy to accelerate convergence and improve the cost-backlog tradeoff without sacrificing asymptotic optimality.

Proposed method

• Proposes a stochastic dual gradient method that decouples decision-making across VMs by using local queue backlogs as dual variables.
• Employs a Lyapunov optimization framework to jointly minimize time-average cost and stabilize service queues, achieving a $[ heta, heta^{-1}]$ cost-backlog tradeoff.
• Introduces a multi-timescale architecture where VNF placement decisions are updated less frequently than service processing, reducing disruption.
• Designs a learn-and-adapt strategy that dynamically adjusts the optimization parameters based on historical data to improve convergence speed.
• Uses a primal-dual update rule where each VM locally computes gradients based on its own backlog and service demand, enabling fully distributed operation.
• Proves asymptotic optimality and stability using Lyapunov drift and perturbation analysis, with rigorous bounds on optimality loss under approximation.

Experimental results

Research questions

• RQ1 How can VNF placement and service processing be optimized in a decentralized, online manner under stochastic service arrivals and dynamic pricing?
• RQ2 What is the achievable cost-backlog tradeoff in a fully distributed NFV system with coupled VM decisions?
• RQ3 Can a multi-timescale design reduce service interruptions from frequent VNF reconfiguration while preserving asymptotic optimality?
• RQ4 How can a learn-and-adapt mechanism improve the convergence speed and tradeoff performance of online NFV optimization?
• RQ5 What is the impact of approximation errors on the optimality gap in the proposed decentralized framework?

Key findings

• The proposed method achieves a cost-backlog tradeoff of $[ heta, heta^{-1}]$, proving asymptotic optimality in minimizing time-average cost while stabilizing service queues.
• The learn-and-adapt strategy improves the tradeoff to $[ heta, heta^{-1} ext{log}^2(1/ heta)]$, significantly enhancing convergence speed without losing optimality.
• Numerical results show a 30% reduction in time-average NFV cost compared to existing benchmarks under dynamic traffic and pricing.
• The method reduces queue length (proxy for delay) by 83% compared to baseline approaches, demonstrating strong performance in congestion control.
• The multi-timescale design effectively reduces service interruptions from VNF reconfiguration, enabling practical deployment in production networks.
• Theoretical analysis confirms that the optimality loss due to approximation is bounded and diminishes with time, validating the robustness of the approach.