On Enhancing Delay SLAs in TCP Networks through Joint Routing and Transport Assistant Deployment

TL;DR

This paper tackles TCP retransmission-induced delays by jointly routing flows and deploying Transport Assistants ($TA$) along the paths. It develops two ILP formulations, $ILP1$ for best-effort delay minimization and $ILP2$ for SLA-aware QoS with deployment costs, and introduces a scalable heuristic, $TAFS$, to handle large networks. Simulations on Abilene, Geant, and Germany show that the joint routing+TA approach reduces end-to-end delay by up to $16.4\%$ and yields substantial cost savings (up to $60.98\%$) under SLA constraints, with near-optimal performance achieved by $TAFS$. Overall, integrating TA placement into routing decisions offers meaningful, practically relevant improvements for TCP performance in both best-effort and QoS-enabled networks.

Abstract

The Transport Control Protocol has long been the primary transport protocol for applications requiring performance and reliability over the Internet. Unfortunately, due its retransmission mechanism, TCP incurs high packet delivery delays when segments are lost. To address this issue, previous research proposed to use a novel network function, namely Transport Assistant, deployed within the network to cache and retransmit lost packets, thus reducing retransmission delays. In this paper, we propose to jointly route the flows and deploy TAs in order to minimize packet delivery delays in best-effort networks (scenario 1) or to satisfy delay-based Service Level Agreements in QoS-based networks (scenario 2). We hence formulate the joint routing and TA deployment problem as Integer Linear Program for the two scenarios and propose a heuristic solution for large-scale instances of the problem. Through extensive simulations, we demonstrate the benefits of performing joint routing flows and TA deployment in reducing packet delivery delays (up to 16.4%) while minimizing deployment costs (up to 60.98%).

Figures (6)

• Figure 1: Packet delivery delay when the packet is lost $a$ times between the source and the TA, and $b$ times between the TA and the destination JaimeTAPlacementNOMS2024.
• Figure 2: Possible paths for communication between nodes h1 and h2 in the Abilene topology.
• Figure 3: Results for Objective 1 (ILP1)
• Figure 4: Results for Objective 2 (ILP2).
• Figure 5: Average EPDD improvement vs. percentage of nodes with TAs for ILP1 and TAFS1.