.png){kind=icon}
.png){kind=icon}
Return
A Single-Order Mixed-Consistency Replication Scheme
.png){kind=icon}
PRE
AI
DOI：10.1145/3772290.3772304.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
We propose Creek, a new protocol that can be used to geo-replicate services which require high availability and have operations with different consistency demands. In our system, replicas can process client requests either in a strongly consistent manner (which simplifies programming, but requires costly synchronization), or an eventually consistent manner (which greatly decreases latency). The unique feature of our protocol is that it progressively totally orders all these requests. In effect, there is no conflicts between concurrent operations, so the service code does not have to resolve any conflicts. Each replica maintains the total order using timestamps propagated asynchronously with reliable broadcast, and conditional atomic broadcast (CAB)-a new total-order protocol, which can also be used separately from Creek. In Creek, CAB ensures linearizable execution of strongly consistent operations, and stabilizes the execution order of causally related weakly consistent operations, each time consensus can be solved. We proved the correctness of CAB and Creek and compared its performance with other closely-related mixed consistency replication protocols.
Keywords:
mixed consistency
geo-replication
total-order broadcast
Journal
P
PROCEEDINGS OF THE 27TH INTERNATIONAL CONFERENCE ON DISTRIBUTED COMPUTING AND NETWORKING, ICDCN 2026
IF:
0
Papers:
21
Citations:
0