Coordination Protocols for Verifiable Consistency in Distributed Storage Systems

Achieving consistency in a highly available distributed storage system has been formally proven to be an impossible task when the system faces network partitions and faulty processes. The complexity is exacerbated when the system allows concurrent processes to send transactions to all the other servers and coordinate the consistent commitment of different transactions. In the event of a partition, each server may allow clients to request updates involving the current state of the data, which makes achieving replicated consistency challenging. To solve the inconsistency problems, several consensus protocols are used, but have strict requirements in order to make progress and are not guaranteed to ever converge to a single value. Additionally, the coordination required to achieve consistency after a partition will be extremely high as each node must compare transaction times and conflicting data with all other servers in the system. To address the inconsistency in distributed systems, this thesis proposes a new coordination protocol that utilizes four ideas in order for clients to verify the consistency of data: (1) a universal timestamp signatory to certify the global order of events, (2) a relative consistency indicator to determine relative consistency during partitions, (3) an operation-based recency-weighted conflict resolution algorithm to simplify coordination for achieving global consistency, and (4) a rejection-oriented distributed transaction commit protocol to eliminate any guarantees required by atomic commit protocols and verify local consistency. This thesis will evaluate and analyze various issues related to coordination and concurrency under network partitions. The experimental results demonstrate that the proposed methods provide a verifiable consistency to the servers of the distributed storage systems