Byzantine Fault Tolerance for Distributed Systems

The growing reliance on online services imposes a high dependability requirement on the computer systems that provide these services. Byzantine fault tolerance (BFT) is a promising technology to solidify such systems for the much needed high dependability. BFT employs redundant copies of the servers and ensures that a replicated system continues providing correct services despite the attacks on a small portion of the system. In this dissertation research, I developed novel algorithms and mechanisms to control various types of application nondeterminism and to ensure the long-term reliability of BFT systems via a migration-based proactive recovery scheme. I also investigated a new approach to significantly improve the overall system throughput by enabling concurrent processing using Software Transactional Memory (STM). Controlling application nondeterminism is essential to achieve strong replica consistency because the BFT technology is based on state-machine replication, which requires deterministic operation of each replica. Proactive recovery is necessary to ensure that the fundamental assumption of using the BFT technology is not violated over long term, i.e., less than one-third of replicas remain correct. Without proactive recovery, more and more replicas will be compromised under continuously attacks, which would render BFT ineffective. STM based concurrent processing maximized the system throughput by utilizing the power of multi-core CPUs while preserving strong replication consistenc