Scalable Byzantine Replication Under Attack

An NSF grant 0716620 (August 2007 - July 2010) to Johns Hopkins University. A component of the NSF Cyber Trust program.
Principal Investigator: Yair Amir.

Overview

As network environments become increasingly hostile, even well protected distributed information systems, constructed with security in mind, are likely to be compromised. Hence, architecting large-scale distributed systems that function correctly and provide adequate performance even when parts of them are compromised is one of the most important challenges. Byzantine replication is emerging as a promising direction to mitigate server compromises. Experience with Byzantine replication protocols reveals considerable shortcomings in the underlying theoretical foundation. Namely, existing fault models, metrics, and correctness criteria used to reason about and construct Byzantine replication algorithms, fail to capture properties that manifest themselves in wide-area environments. Since all existing Byzantine replication algorithms were designed to meet the standard safety and liveness criteria, they all exhibit critical vulnerabilities not covered by the standard models. This project will develop the theoretical foundation, architectural framework, and algorithmic techniques for a scalable wide-area Byzantine replication system that provides strong performance guarantees under attack. This includes:

  • Expanding the existing theoretical fault models to better encapsulate the unique characteristics and performance vulnerabilities associated with scalable wide-area Byzantine replication systems.
  • Defining useful metrics for evaluating and comparing different architectures, configurations, and algorithms with a focus on their performance under sophisticated attacks.
  • Developing an architectural framework for scalable Byzantine replication that can be customized to topology, performance, and resiliency requirements of specific wide-area systems.
  • Developing specific algorithms for wide-area environments that provide strong performance guarantees under attack.

Students

Results and Current Activities

  • We have shown that standard leader-based replication protocols are vulnerable to performance attacks that can render them useless in practice even though they meet the traditional correctness criteria of safety and liveness.
     
  • We have devised a new correctness criterion, Bounded-Delay, to assess the performance of state machine replication protocols when under attack.
     
  • We developed Prime (Performance-Oriented Replication In Malicious Environments), a new state machine replication protocol that achieves a new level of resilience to performance attacks and meets the Bounded-Delay correctness criterion.
     
  • Collaborating with the Navigators group while at the University of Lisboa, Portugal, we analyzed the extent to which randomized intrusion-tolerant replication protocols (which do not rely on a leader for coordination) are vulnerable to performance degradation under attack. We devised a timing attack that can likely be used to significantly reduce performance.
     
  • We developed a new view change protocol for Prime that offers a fundamentally stronger performance guarantee by preventing faulty servers from delaying progress by causing long view changes.
     
  • We designed and are implementing a scalable attack-resilient architecture for intrusion-tolerant wide-area replication.

Related Publications

  • Byzantine Replication Under Attack
    Yair Amir, Brian Coan, Jonathan Kirsch, John Lane
    In Proceedings of the 38th IEEE/IFIP International Conference on Dependable Systems and Networks (DSN 2008), Anchorage, Alaska, June 2008, pp. 197-206.

    Byzantine-resilient replication protocols satisfy two standard correctness criteria, safety and liveness, in the presence of Byzantine faults. In practice, however, faulty processors can, in some protocols, significantly degrade performance by causing the system to make progress at an extremely slow rate. While ``correct'' in the traditional sense, systems vulnerable to such performance degradation are of limited practical use in adversarial environments. This paper argues that techniques for mitigating such performance attacks are needed to bridge this ``practicality gap'' for intrusion-tolerant replication systems. We propose a new performance-oriented correctness criterion, and we show how failure to meet this criterion can lead to performance degradation. We present a new Byzantine replication protocol that achieves the criterion and evaluate its performance in fault-free configurations and when under attack.


  • Prime: Byzantine Replication Under Attack
    Yair Amir, Brian Coan, Jonathan Kirsch, John Lane
    Technical Report CNDS-2009-4.

Presentations