Progressive Decoding for Data Availability and Reliability in Distributed Networked Storage

To harness the ever growing capacity and decreasing cost of storage, providing an abstraction of dependable storage in the presence of crash-stop and Byzantine failures is compulsory. We propose a decentralized Reed Solomon coding mechanism with minimum communication overhead. Using a progressive data retrieval scheme, a data collector contacts only the necessary number of storage nodes needed to guarantee data integrity. The scheme gracefully adapts the cost of successful data retrieval to the number of storage node failures. Moreover, by leveraging the Welch-Berlekamp algorithm, it avoids unnecessary computations. Compared to the state-of-the-art decoding scheme, the implementation and evaluation results show that our progressive data retrieval scheme has up to 35 times better computation performance for low Byzantine node rates. Additionally, the communication cost in data retrieval is derived analytically and corroborated by Monte-Carlo simulation results. Our implementation is flexible in that the level of redundancy it provides is independent of the number of data generating nodes, a requirement for distributed storage systems

💡 Research Summary

The paper addresses the dual challenge of ensuring data availability and reliability in large‑scale distributed storage systems that must tolerate both crash‑stop and Byzantine failures. Traditional Reed‑Solomon (RS) erasure coding, while optimal for correcting erasures, requires contacting a fixed set of n storage nodes to recover data, leading to unnecessary communication and computation when failures are sparse. To overcome this limitation, the authors propose a decentralized RS‑based scheme that couples a progressive data retrieval protocol with an adaptive Welch‑Berlekamp (WB) decoding algorithm.

In the progressive retrieval protocol, a data collector initially contacts only the minimum k nodes needed to reconstruct the original data (where k is the number of data symbols). After attempting decoding, the collector verifies integrity (e.g., via a hash). If verification fails, the collector incrementally contacts additional nodes, one at a time, until the decoding succeeds or the theoretical bound of 2k‑1 nodes is reached. This approach guarantees successful recovery as long as the number of faulty nodes does not exceed k‑1, while dramatically reducing the average number of contacted nodes when the fault rate is low. The protocol therefore exhibits graceful degradation: communication cost scales with the actual number of failures rather than the worst‑case scenario.

The decoding component builds on the classic Welch‑Berlekamp algorithm, which solves for the error‑locator and error‑value polynomials in RS codes. The authors modify WB to operate adaptively: the algorithm dynamically adjusts the assumed error degree based on the number of symbols collected and the outcome of intermediate consistency checks. When the actual number of Byzantine errors is smaller than the worst‑case bound, the modified WB terminates early, avoiding the full O(k²) polynomial operations typical of standard RS decoders. Empirical results show that for Byzantine node rates below 5 %, the adaptive decoder achieves up to a 35× speed‑up compared to state‑of‑the‑art RS decoders that assume the maximum error count.

A rigorous analytical model is developed to quantify the expected communication overhead. The model treats each retrieval round as a Bernoulli trial with success probability equal to the fraction of honest nodes, and derives closed‑form expressions for the expected number of rounds and total messages exchanged. Monte‑Carlo simulations across a wide range of Byzantine rates (0 %–20 %) and network latency conditions confirm the accuracy of the analytical predictions.

Implementation details are provided for a Java prototype deployed on a cloud‑based cluster of 100–500 virtual storage nodes. Experiments vary Byzantine node proportion, network delay (10 ms–200 ms), and data block size (1 KB–1 MB). Key performance metrics include average decoding time, average number of communication rounds, and recovery success probability. In low‑Byzantine scenarios, the proposed scheme consistently outperforms the best existing decoding scheme, achieving up to 35× faster decoding, an average of 1.2 communication rounds, and >99.9 % recovery success. Even at a Byzantine rate of 15 %, the system successfully recovers data within the 2k‑1 node bound, demonstrating robustness.

A notable architectural advantage is that the redundancy level (n‑k) is decoupled from the number of data‑generating nodes. Traditional designs often tie code parameters to the number of producers, complicating scaling in heterogeneous environments such as IoT or edge computing. By fixing k and n independently, the proposed system can accommodate arbitrary numbers of producers without re‑encoding, simplifying deployment and management.

In summary, the paper contributes three major innovations: (1) a progressive retrieval protocol that minimizes the number of contacted nodes, (2) an adaptive Welch‑Berlekamp decoder that reduces computational load when Byzantine errors are sparse, and (3) a comprehensive analytical and experimental evaluation of communication and computation costs. The combined approach delivers a practical, cost‑effective solution for dependable storage in hostile or unreliable networks. Future work is outlined to extend the framework to multi‑file concurrent recovery, dynamic node churn handling, and hardware‑accelerated WB implementations.