[Paper] SCRamble: Adaptive Decentralized Overlay Construction for Blockchain Networks

Source: arXiv - 2601.10277v1

Overview

The paper presents SCRamble, a decentralized overlay‑construction protocol that reshapes how blockchain nodes choose their peers. By replacing the traditional “random‑peer” approach with a latency‑aware, performance‑driven link selection, SCRamble cuts block propagation time dramatically, which in turn lifts transaction throughput and reduces fork risk.

Key Contributions

• Adaptive Peer‑Selection Heuristics – Introduces two complementary scoring mechanisms: (1) arrival‑time scoring that rewards neighbors delivering blocks quickly, and (2) latency‑aware scoring that prefers low‑delay links.
• Decentralized Overlay Construction – Implements the heuristics in a fully peer‑to‑peer fashion without any central coordinator or global view of the network.
• Analytical Model & Guarantees – Provides a formal model showing how the combined scores converge to a low‑diameter, high‑bandwidth overlay.
• Empirical Evaluation – Demonstrates up to 45 % reduction in block dissemination latency and a 30 % increase in transaction throughput on both synthetic and real‑world blockchain testbeds.
• Security Insight – Shows that faster propagation directly lowers the probability of temporary forks, strengthening consensus security.

Methodology

1. Scoring Functions

   • Arrival‑Time Score: Each node tracks how soon it receives a newly mined block from each neighbor, normalizing over recent blocks to smooth out outliers.
   • Latency Score: Nodes periodically ping peers to estimate round‑trip time (RTT). The score is inversely proportional to RTT, encouraging geographically or topologically close connections.

2. Peer‑Selection Algorithm

   • Nodes maintain a bounded neighbor table (e.g., 8–12 peers).
   • Periodically (every few minutes) they compute a combined score S = α·A + (1‑α)·L (where A is arrival‑time score, L is latency score, and α tunes the trade‑off).
   • The lowest‑scoring peers are dropped and replaced by randomly discovered candidates, which are immediately evaluated and possibly adopted.

3. Simulation & Testbed Setup

   • Synthetic Networks: Random geometric graphs with 1 k–10 k nodes, varying latency distributions to emulate global Internet conditions.
   • Real‑World Testbed: A fork of the Ethereum Go client (Geth) running on 200 nodes across multiple cloud regions (AWS, GCP, Azure).
   • Metrics collected: block propagation latency (time from block creation to 90 % of nodes receiving it), transaction throughput (tps), and fork rate.

4. Baseline Comparisons

   • Standard random‑peer overlay (as used in Bitcoin/Ethereum).
   • Prior latency‑aware approaches (e.g., “Closest‑Peers” heuristics).

Results & Findings

• Latency Reduction: SCRamble consistently outperforms both baselines, achieving up to a 45 % cut in median propagation time.
• Throughput Boost: Faster dissemination translates into higher sustainable transaction rates, with a 30 % increase over the random baseline.
• Stability: The overlay quickly stabilizes after churn events (node joins/leaves), typically within two re‑selection cycles (~10 minutes).
• Security Benefit: The measured fork probability drops by more than half, indicating a stronger consensus under the same mining power.

Practical Implications

• Higher TPS for Existing Chains – Deploying SCRamble as a drop‑in overlay replacement could lift the practical transaction capacity of Bitcoin‑like or Ethereum‑like networks without protocol‑level changes.
• Reduced Confirmation Times – Faster block spread means users see confirmations sooner, improving UX for payment and DeFi applications.
• Lower Bandwidth Costs – By pruning high‑latency links, nodes send fewer duplicate messages, saving on cloud egress fees for hosted validators.
• Improved Security Posture – Lower fork rates diminish the attack surface for selfish‑mining or double‑spend attempts, a compelling argument for validators and exchanges.
• Ease of Integration – SCRamble operates purely at the networking layer; existing client implementations can adopt it by swapping the peer‑selection module, making it attractive for open‑source projects.

Limitations & Future Work

• Parameter Sensitivity – The α weighting between arrival‑time and latency scores needs tuning per network topology; a sub‑optimal α can degrade performance.
• Bootstrapping Overhead – New nodes initially rely on random peers before enough data is gathered to compute meaningful scores, causing a short “cold‑start” latency spike.
• Adversarial Manipulation – Malicious peers could falsify latency probes or artificially delay blocks to appear more “useful”; the authors suggest cryptographic timestamp verification as a mitigation.
• Scalability to Very Large Networks – Experiments capped at 10 k nodes; further work is needed to confirm behavior at Bitcoin‑scale (≈ 10 M nodes).

Future directions include adaptive α learning (e.g., reinforcement learning), integration with gossip‑based protocols for even faster dissemination, and formal security analysis against targeted latency attacks.

SCRamble shows that smarter, decentralized networking can unlock tangible performance gains for blockchains today, offering a pragmatic path for developers and operators to boost throughput and security without overhauling consensus logic.

Authors

• Evangelos Kolyvas
• Alexandros Antonov
• Spyros Voulgaris

Paper Information

• arXiv ID: 2601.10277v1
• Categories: cs.DC, cs.NI
• Published: January 15, 2026
• PDF: Download PDF