[Paper Review] M2M Traffic via Random Access Satellite links: Interactions between Transport and MAC Layers

Manlio Bacco, Tomaso de Cola+2 more|arXiv (Cornell University)|Sep 12, 2016

IoT Networks and Protocols17 references3 citations

TL;DR

This paper proposes a novel analytical model, NewRenoSAT, to accurately estimate TCP NewReno throughput over random access satellite links using the CRDSA++ MAC protocol (3 replicas) in DVB-RCS2 networks. It demonstrates that TCP congestion control inherently stabilizes access load without centralized control, and identifies optimal waveform and MSS configurations for M2M traffic under bursty, low-rate conditions.

ABSTRACT

Machine-to-machine services are witnessing an unprecedented diffusion, which is expected to result in an ever-increasing data traffic load. In this context, satellite technology is playing a pivotal role, since it enables a widespread provisioning of machine-to-machine services. In particular, oil industry, maritime communications, as well as remote monitoring are sectors where the use of satellite communications is expected to dramatically explode within the next few years. In the light of this sudden increase of machine-to-machine data transported over satellite, a more thorough understanding of machine-to-machine service implementation over satellite is required, especially focusing on the interaction between transport and MAC layers of the protocol stack. Starting from these observations, this paper thoroughly analyses the interaction between TCP and the Contention Resolution Diversity Slotted Aloha access scheme defined in the DVB-RCS2 standard, assuming the use of an MQTT-like protocol to distribute machine-to-machine services. A novel TCP model is developed and validated through extensive simulation campaigns, which also shed important lights on the design choices enabling the efficient transport of machine-to-machine data via satellite.

Motivation & Objective

• To address the lack of rigorous analytical models linking TCP transport layer performance with random access MAC protocols in satellite M2M communications.
• To investigate how TCP congestion control interacts with collision resolution in CRDSA++-based satellite access, particularly under bursty M2M traffic.
• To validate the proposed model through NS-3 simulations and identify optimal system configurations (waveform, MSS) for efficient M2M data transport via satellite.
• To determine whether TCP's inherent congestion control can replace centralized load control mechanisms in DVB-RCS2 networks.
• To develop a bridge model linking MAC-layer burst losses (BLR) to transport-layer segment loss rates (q) for accurate throughput prediction.

Proposed method

• Develops a new analytical TCP model, NewRenoSAT, adapted from [11–13], to model TCP dynamics over random access channels with collision-based losses.
• Introduces a BLR (Burst Loss Rate) model to map MAC-layer collisions (due to CRDSA++ contention) to transport-layer segment loss rates, enabling accurate loss event estimation.
• Uses NS-3-based simulations to validate the NewRenoSAT model under varying numbers of RCSTs, waveforms (WF 14, WF 3), and MSS sizes.
• Employs a load control analysis comparing the equilibrium offered load $\hat{G}_T$ under TCP to the optimal load $G^*$ of CRDSA++ without TCP, showing TCP’s self-stabilizing behavior.
• Evaluates performance across different configurations (e.g., 64 vs. 194 time-slots per RA block) to identify optimal waveform and MSS combinations.
• Applies the model to estimate throughput and loss rates under realistic M2M traffic patterns, assuming MQTT-like application layer behavior.

Experimental results

Research questions

• RQ1 How does TCP NewReno perform over a CRDSA++-based random access satellite link under varying M2M traffic loads?
• RQ2 To what extent can TCP congestion control stabilize the access load without centralized load control mechanisms in DVB-RCS2 networks?
• RQ3 What is the relationship between MAC-layer burst losses (BLR) and transport-layer segment loss rates (q) in random access satellite systems?
• RQ4 Which waveform (WF 14 vs. WF 3) and MSS size combination maximizes throughput and resource utilization for M2M satellite communications?
• RQ5 How does the number of RCSTs affect the optimal system configuration and performance in terms of offered load per station and aggregate throughput?

Key findings

• The NewRenoSAT model achieves significantly lower estimation error than existing models, with relative errors below 0.07 across tested configurations.
• TCP NewReno stabilizes the system at $\hat{G}_T \approx 0.45–0.55$ (WF 14) and $\hat{G}_T \approx 0.58–0.63$ (WF 3), which are below the optimal $G^*$ of 0.7 and 0.78, respectively, due to congestion control feedback.
• For up to ~100 RCSTs, WF 14 (64 time-slots) provides better performance than WF 3 (194 time-slots), but for larger populations, WF 3 is superior due to finer time-slot granularity.
• Configurations leading to $f > 1$ (fragmentation of TCP segments at MAC layer) result in poor performance and should be avoided, as shown by low throughput and high sub-utilization.
• The BLR model accurately maps MAC-level collisions to transport-level losses, with a close match to simulation results, validating its use in the NewRenoSAT framework.
• TCP’s distributed congestion control mechanism effectively replaces the need for complex centralized load control in DVB-RCS2, ensuring stability and efficient resource use without additional signaling.