Safe and secure distributed computing

ACES team

The objective is to establish reliable data access and consensus protocols that optimize energy consumption, throughput, and latency, while ensuring robustness against Byzantine attacks and improving data protection techniques.

🔗Team activity between 2018 and 2023

The following text has been written by the ACES team as part of the 2018-2023 periodic HCÉRES evaluation of the LTCI lab and reflects the past activities of the team on the "Safe and secure distributed computing" topic.

Today's computing systems, ranging from fire-alarm sensors and laptops to internet-scale services, are inherently distributed. Our research delves into the fundamentals and system aspects of distributed computing under fault-prone and loosely synchronous models. Addressing security, particularly against Byzantine attacks where components may be fully compromised, is vital in open, large-scale systems. Blockchain emerges as a solution for reliable data access amidst mutual distrust, leveraging protocols such as reliable broadcast, consensus, state-machine replication, erasure coding, and zero-knowledge proofs.

We investigate the inherent drawbacks of these protocols –energy consumption, throughput, and latency– and propose methods to overcome them. Specifically, we explore the cryptocurrency (asset-transfer) challenge, aiming for efficient solutions that reduce the need for global synchronization. Our research has led to innovations like a fast Byzantine consensus algorithm requiring only 5f-1 replicas ("Revisiting Optimal Resilience of Fast Byzantine Consensus"), accountable lattice agreement protocol ("Brief Announcement: Accountability and Reconfiguration -Self-Healing Lattice Agreement"), scalable reliable broadcast protocol ("Scalable Byzantine Reliable Broadcast"), a consensusless cryptocurrency prototype ("Permissionless and Asynchronous Asset Transfer"), comparative analyses of blockchains and Byzantine fault-tolerant protocols ("The devil hides in the model: Reviewing Blockchain and BFT protocols"), and a concurrency-optimal solution for asynchronous asset transfer ("CryptoConcurrency: (Almost) Consensusless Asset Transfer with Shared Accounts").

In conjunction with its focus on safety and security in distributed systems, our team also prioritizes data protection, particularly for images, against prying storage providers. Techniques include data fragmentation, encryption, and dispersion, optimizing classic secret sharing for better performance and memory efficiency with some security compromise ("Enhancing data protection with a structure-wise fragmentation and dispersal of encrypted data", "All-Or-Nothing data protection for ubiquitous communication: Challenges and perspectives"). Image protection strategies involve frequency domain fragmentation and novel algorithms for reconstruction without prior knowledge ("DC coefficient recovery for JPEG images in ubiquitous communication systems"). Moreover, deep learning enhances these methods and proves effective in breaching systems, exemplified by attacking a network intrusion detection system with minor dataset modifications yet high success rates ("Adversarial Attacks Against Network Intrusion Detection in IoT Systems").