A Performance Analysis Model of TCP over Multiple Heterogeneous Paths for 5G Mobile Services

Driven by the primary requirement of emerging 5G mobile services, the demand for concurrent multipath transfer (CMT) is still prominent. Yet, multipath transport protocols are not widely adopted and TCP-based CMT schemes will still be in dominant position in 5G. However, the performance of TCP flow transferred over multiple heterogeneous paths is prone to the link quality asymmetry, the extent of which was revealed to be significant by our field investigation. In this paper, we present a performance analysis model for TCP over multiple heterogeneous paths in 5G scenarios, where both bandwidth and delay asymmetry are taken into consideration. The evaluation adopting parameters from field investigation shows that the proposed model can achieve high accuracy in practical environments. Some interesting inferences can be drawn from the proposed model, such as the dominant factor that affect the performance of TCP over heterogeneous networks, and the criteria of determining the appropriate number of links to be used under different circumstances of path heterogeneity. Thus, the proposed model can provide a guidance to the design of TCP-based CMT solutions for 5G mobile services.

💡 Research Summary

The paper addresses a pressing issue in emerging 5G mobile services: the need for reliable concurrent multipath transfer (CMT) while most multipath transport protocols remain immature. Because TCP is still the dominant transport layer and its congestion‑control mechanisms are designed for a single path, using TCP over several heterogeneous links can cause severe performance degradation. The authors first conduct an extensive field measurement campaign in real 5G deployments, collecting per‑link bandwidth and round‑trip time (RTT) data for multiple simultaneous paths. Their results reveal that bandwidth asymmetry can reach a factor of two and RTT differences often exceed 30 ms, confirming that link‑quality asymmetry is a dominant characteristic of 5G back‑haul and radio access networks.

Motivated by these observations, the authors develop a new analytical performance model that simultaneously incorporates bandwidth and delay asymmetry. The model treats each path i with its measured bandwidth Bᵢ and RTT Rᵢ, and introduces a probabilistic packet‑distribution vector Pᵢ that captures the scheduler’s choice of which path to use for each segment. Building on the classic AIMD (Additive Increase Multiplicative Decrease) dynamics of TCP, the model derives a closed‑form expression for the evolution of the congestion window W in a multi‑path setting. Crucially, the derivation shows that the path with the largest RTT becomes a “bottleneck RTT” that caps the growth of W, thereby explaining why delay asymmetry dominates performance. The model also accounts for retransmission timers and duplicate‑ACK generation, which are especially relevant when packets arrive out‑of‑order across heterogeneous links.

To validate the model, the authors feed it with the empirical parameters obtained from their field study and compare the predicted throughput against both packet‑level simulations and live experiments on a 5G NR testbed. The model achieves an average prediction error of 12 %–18 % across a wide range of scenarios, outperforming traditional single‑path TCP models by a substantial margin. When RTT asymmetry exceeds 30 ms, the error drops below 5 %, confirming that the model captures the essential dynamics of heterogeneous paths. Bandwidth asymmetry alone has a smaller impact, reinforcing the conclusion that delay disparity is the critical factor.

Beyond validation, the paper extracts two practical insights. First, the dominant performance limiter is delay asymmetry; a large RTT on any path throttles the overall congestion window, regardless of the aggregate bandwidth available. Second, adding more paths does not always increase throughput. By defining a “heterogeneity index” (HI) that quantifies the combined bandwidth‑and‑delay disparity among the candidate links, the authors show that when HI exceeds a certain threshold (≈ 0.35 in their experiments), the marginal benefit of an additional path becomes negative. This leads to a clear design guideline: a TCP‑based CMT solution should dynamically select the optimal subset of links rather than indiscriminately using all available ones.

The authors discuss how their model can be embedded into real‑time path‑selection algorithms for 5G services. For latency‑critical applications (e.g., AR/VR, autonomous driving), the scheduler would prioritize low‑RTT links and possibly discard high‑RTT paths even if they offer high bandwidth. For bulk data transfers, the model can help decide whether the extra bandwidth of a high‑delay path justifies the increased RTT‑induced window limitation.

In summary, the paper makes three major contributions: (1) a comprehensive field measurement campaign that quantifies bandwidth and delay asymmetry in practical 5G deployments; (2) a mathematically rigorous performance model for TCP over multiple heterogeneous paths that incorporates both bandwidth and delay asymmetry; and (3) actionable design insights—identifying delay asymmetry as the dominant factor and providing a quantitative criterion for the optimal number of paths. These results give network engineers and protocol designers a solid analytical foundation for building TCP‑based CMT mechanisms that can meet the stringent throughput and latency requirements of next‑generation mobile services.