arXiv subjects

Antoine Desjardins

Publications and source records attributed to Antoine Desjardins.

4 recordsLinked to original sources

arXiv2025

RACS-SADL: Robust and Understandable Randomized Consensus in the Cloud

Widely deployed consensus protocols in the cloud are often leader-based and optimized for low latency under synchronous network conditions. However, cloud networks can experience disruptions such as network partitions, high-loss links, and configuration errors. These disruptions interfere with the operation of leader-based protocols, as their view change mechanisms interrupt the normal case replication and cause the system to stall. We propose RACS, a novel randomized consensus protocol that ensures robustness against adversarial network conditions. RACS achieves optimal one-round trip latency under synchronous network conditions while remaining resilient to adversarial network conditions. RACS follows a simple design inspired by Raft, the most widely used consensus protocol in the cloud, and therefore enables seamless integration with the existing cloud software stack. Experiments with a prototype running on Amazon EC2 show that RACS achieves 28k cmd/sec throughput, ninefold higher than Raft under adversarial cloud network conditions. Under synchronous network conditions, RACS matches the performance of Multi-Paxos and Raft, achieving a throughput of 200k cmd/sec with a median latency of 300ms, confirming that RACS introduces no unnecessary overhead. Finally, SADL-RACS, a throughput-optimized version of RACS, achieves a throughput of 500k cmd/sec, delivering 150% higher throughput than Raft.

cs.DC↗

arXiv2023

Constrained Adaptive Attacks: Realistic Evaluation of Adversarial Examples and Robust Training of Deep Neural Networks for Tabular Data

State-of-the-art deep learning models for tabular data have recently achieved acceptable performance to be deployed in industrial settings. However, the robustness of these models remains scarcely explored. Contrary to computer vision, there is to date no realistic protocol to properly evaluate the adversarial robustness of deep tabular models due to intrinsic properties of tabular data such as categorical features, immutability, and feature relationship constraints. To fill this gap, we propose CAA, the first efficient evasion attack for constrained tabular deep learning models. CAA is an iterative parameter-free attack that combines gradient and search attacks to generate adversarial examples under constraints. We leverage CAA to build a benchmark of deep tabular models across three popular use cases: credit scoring, phishing and botnet attacks detection. Our benchmark supports ten threat models with increasing capabilities of the attacker, and reflects real-world attack scenarios for each use case. Overall, our results demonstrate how domain knowledge, adversarial training, and attack budgets impact the robustness assessment of deep tabular models and provide security practitioners with a set of recommendations to improve the robustness of deep tabular models against various evasion attack scenarios.

cs.LG↗

arXiv2023

Divide and Scale: Formalization and Roadmap to Robust Sharding

Sharding distributed ledgers is a promising on-chain solution for scaling blockchains but lacks formal grounds, nurturing skepticism on whether such complex systems can scale blockchains securely. We fill this gap by introducing the first formal framework as well as a roadmap to robust sharding. In particular, we first define the properties sharded distributed ledgers should fulfill. We build upon and extend the Bitcoin backbone protocol by defining consistency and scalability. Consistency encompasses the need for atomic execution of cross-shard transactions to preserve safety, whereas scalability encapsulates the speedup a sharded system can gain in comparison to a non-sharded system. Using our model, we explore the limitations of sharding. We show that a sharded ledger with $n$ participants cannot scale under a fully adaptive adversary, but it can scale up to $m$ shards where $n=c'm\log m$, under an epoch-adaptive adversary; the constant $c'$ encompasses the trade-off between security and scalability. This is possible only if the sharded ledgers create succinct proofs of the valid state updates at every epoch. We leverage our results to identify the sufficient components for robust sharding, which we incorporate in a protocol abstraction termed Divide & Scale. To demonstrate the power of our framework, we analyze the most prominent sharded blockchains (Elastico, Monoxide, OmniLedger, RapidChain) and pinpoint where they fail to meet the desired properties.

cs.DC↗

arXiv2022

Mandator and Sporades: Robust Wide-Area Consensus with Efficient Request Dissemination

Consensus algorithms are deployed in the wide area to achieve high availability for geographically replicated applications. Wide-area consensus is challenging due to two main reasons: (1) low throughput due to the high latency overhead of client request dissemination and (2) network asynchrony that causes consensus protocols to lose liveness. In this paper, we propose Mandator and Sporades, a modular state machine replication algorithm that enables high performance and resiliency in the wide-area setting. To address the high client request dissemination overhead challenge, we propose Mandator, a novel consensus-agnostic asynchronous dissemination layer. Mandator separates client request dissemination from the critical path of consensus to obtain high performance. Composing Mandator with Multi-Paxos (Mandator-Paxos) delivers significantly high throughput under synchronous networks. However, under asynchronous network conditions, Mandator-Paxos loses liveness which results in high latency. To achieve low latency and robustness under asynchrony, we propose Sporades, a novel omission fault-tolerant consensus algorithm. Sporades consists of two modes of operations -- synchronous and asynchronous -- that always ensure liveness. The combination of Mandator and Sporades (Mandator-Sporades) provides a robust and high-performing state machine replication system. We implement and evaluate Mandator-Sporades in a wide-area deployment running on Amazon EC2. Our evaluation shows that in the synchronous execution, Mandator-Sporades achieves 300k tx/sec throughput in less than 900ms latency, outperforming Multi-Paxos, EPaxos and Rabia by 650\% in throughput, at a modest expense of latency. Furthermore, we show that Mandator-Sporades outperforms Mandator-Paxos, Multi-Paxos, and EPaxos in the face of targeted distributed denial-of-service attacks.

cs.DC↗