Distributed Systems Making Byzantine Fault-Tolerant Systems - - PowerPoint PPT Presentation

Making Byzantine Fault-Tolerant Systems Tolerate Byzantine Faults

Hubert Jaworski

Distributed Systems

Byzantine system

 Copies of critical components  Concurent replicas  Response agreement  Omission failure proof  Robustness

Byzantine failure

 Incorrectly working components  State corruption  Inconsistent responses

Byzantine fault tolerance

replication systems:

 PBFT  Q/U  HQ  Zyzzyva  Aardvark

Basic concept

 Accepting requests by primary  Broadcasting requests to replicas  Processing request by replicas  Replica-to-replica communication  Replying to client  Reaching consensus

PBFT

 Practical Byzantine Fault Tolerance  Optimized gracious execution path  Message authentication codes  3-round quorum-based  Consensus over 3f+1 replicas

Connection concept example

PBFT communication patern

Q/U

 Conflict-free Quorum-based BFT Protocol  Query/Update  Single-round quorum-based  Requires 5f + 1 replicas  Carefully optimized  Requires a significantly lower number of

messages

 No primary replica

HQ

 Two-round quorum protocol  Requires 3f + 1 replicas  Quorum of 2f + 1  Better at resolving concurrency conflicts  No primary replica

Zyzzyva

 Speculative Byzantine Fault Tolerance  3f+1 total replicas, 2f+1 replicas with state  MAC authentication  Agreement, view change, checkpoint

protocols

Aardvark

 Robust BFT  Requires 3f+1 replicas  Gracious and uncivil executions  Traded some performace for robustness  Client-side signature generation  hybrid MAC-signature construct  separate network interface controllers  Regular view changes

Aardvark

 Blacklisting  Caching responses  Round-robin client-based request scheduler 

Efficiency

Verifying client request

Message from replica

Faulty client

Faulty replica

Faulty primary

The End

Thank you for your time