[Paper Review] A Distributed Algorithm for Throughput Optimal Routing in Overlay Networks

Anurag Rai, Rahul Singh+1 more|arXiv (Cornell University)|Dec 16, 2016

Peer-to-Peer Network Technologies4 citations

TL;DR

This paper proposes the Optimal Overlay Routing Policy (OORP), a distributed, throughput-optimal algorithm for overlay networks that leverages underlay queue-lengths to dynamically route traffic and avoid instability. OORP outperforms prior heuristics like OBP by accurately accounting for underlay congestion, achieving optimal throughput even with limited visibility into the legacy network.

ABSTRACT

We address the problem of optimal routing in overlay networks. An overlay network is constructed by adding new overlay nodes on top of a legacy network. The overlay nodes are capable of implementing any dynamic routing policy, however, the legacy underlay has a fixed, single path routing scheme and uses a simple work-conserving forwarding policy. Moreover, the underlay routes are pre-determined and unknown to the overlay network. The overlay network can increase the achievable throughput of the underlay by using multiple routes, which consist of direct routes and indirect routes through other overlay nodes. We develop a throughput optimal dynamic routing algorithm for such overlay networks called the Optimal Overlay Routing Policy (OORP). OORP is derived using the classical dual subgradient descent method, and it can be implemented in a distributed manner. We show that the underlay queue-lengths can be used as a substitute for the dual variables. We also propose various schemes to gather the information about the underlay that is required by OORP and compare their performance via extensive simulations.

Motivation & Objective

• To address the suboptimality of existing overlay routing policies like OBP, which fail to account for underlay congestion and can destabilize queues.
• To design a distributed, throughput-optimal routing policy for overlay networks built on legacy underlay networks with fixed single-path routing.
• To enable the overlay to achieve maximum possible throughput by using underlay queue-lengths as proxies for dual variables in a subgradient optimization framework.
• To develop practical methods for estimating underlay congestion when direct queue-length information is unavailable.
• To integrate the routing policy with a rate controller that maximizes network utility under capacity and background traffic constraints.

Proposed method

• Derives a centralized, throughput-optimal policy using dual subgradient descent, where underlay queue-lengths serve as dual variables to limit injected traffic and ensure stability.
• Develops a distributed implementation of the policy where each overlay node uses local underlay queue-length information to compute its routing decisions.
• Proposes three estimation schemes for underlay congestion: probing with control packets, delay-based estimation, and a hybrid approach combining both.
• Introduces a rate control mechanism that adjusts commodity rates to maximize a utility function while respecting link capacities and background traffic.
• Uses a tunnel-based overlay model where traffic can be routed directly or via intermediate overlay nodes, with all decisions based on real-time queue backlog feedback.
• Validates the approach via extensive simulations on various topologies, comparing OORP against OBP and estimation-based variants.

Experimental results

Research questions

• RQ1 Is the Overlay Backpressure Policy (OBP) truly throughput optimal in overlay networks with legacy underlay constraints?
• RQ2 Can a distributed routing policy achieve throughput optimality by using underlay queue-lengths as congestion indicators?
• RQ3 How accurately can underlay congestion be estimated when direct queue-length access is unavailable?
• RQ4 Can the proposed policy be extended to support utility-maximizing rate control under shared link constraints?
• RQ5 What is the performance gap between using actual queue-lengths and estimated congestion measures in real-world deployment scenarios?

Key findings

• The paper provides a counterexample showing that OBP is not throughput optimal, as it can cause instability in underlay queues due to inaccurate congestion awareness.
• OORP achieves throughput optimality by using underlay queue-lengths as proxies for dual variables, ensuring that traffic injection respects underlay capacity and maintains queue stability.
• Probing-based congestion estimation achieves near-optimal performance and significantly outperforms delay-based estimation, especially under high backlog conditions.
• When background traffic is present, OORP maintains throughput optimality by effectively treating it as a reduction in available capacity, preserving stability and performance.
• With rate control, OORP converges to a throughput vector that maximizes network utility, such as [1, 1.3, 0.5], even when individual commodity throughputs are constrained by shared resources.
• Simulations confirm that OORP outperforms OBP and that probing-based estimation closely matches the performance of using actual queue-lengths, validating its practicality.