Back
2026INF001.pdf
Doctoral thesis

Revving up replication

  • 2026

PhD: Università della Svizzera italiana

State machine replication
Fault tolerance
Dependability
English State Machine Replication (SMR) is a fundamental approach for building fault-tolerant distributed systems, ensuring that multiple replicas maintain a consistent state despite failures. However, achieving both strong consistency and high performance remains challenging due to the overhead of replica coordination and state synchronization. This thesis addresses these challenges by proposing novel communication and state management mechanisms for SMR systems. We present FlexCast, an overlay-based and genuine atomic multicast protocol that combines geographical locality with efficient message delivery through a complete DAG overlay. It reduces latency in geographically distributed deployments and eliminates communication overhead. Moreover, we introduce quiescence, our proposed novel atomic multicast property that complements minimality by ensuring that processes cease communication once they are no longer involved in any multicast, thereby reinforcing both efficiency and genuineness. Additionally, we introduce a reconfiguration mechanism to FlexCast, that dynamically adapts its overlay to workload locality, maintaining high performance under changing network conditions. We also address efficient state transfer and verification in replicated systems. We propose two different self-validating clustered data structures that enable efficient, parallel, and independently verifiable state synchronization, reduce tail latencies, improve checkpointing and recovery, and remain robust under Byzantine faults. Integration into a practical SMR framework demonstrates their effectiveness across normal and adversarial scenarios in wide-area networks. Overall, this thesis provides a comprehensive set of techniques to improve the performance, scalability, and fault tolerance of SMR systems, combining efficient communication, dynamic adaptation, and advanced state management to meet the demands of modern distributed applications.
Collections
Language
  • English
Classification
Computer science and technology
License
License undefined
Open access status
green
Identifiers
Persistent URL
https://n2t.net/ark:/12658/srd1334927
Statistics
Document views: 0 File downloads:
  • 2026INF001.pdf: 0