Heterogeneous Quorum Architecture for Two-Data-Center Consensus
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Two-data center storage systems face challenges in achieving quorum-based writes due to the need for an odd number of nodes, leading to increased operational costs and split-brain scenarios, as achieving a majority commitment is not possible with an even number of nodes across data centers.
Innovation Solution
Implementing a heterogeneous quorum-based system with an odd number of participants across multiple availability zones, where some nodes are connected via low-latency high-bandwidth networks and others via higher latency lower-bandwidth networks, ensuring quorum is maintained even if one node fails, using a consensus ring protocol like RAFT or Paxos.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If an even number of nodes is used across two data centers, then operational costs are reduced and simplicity is improved, but quorum-based writes cannot be achieved and split-brain scenarios occur
Solution Approach 1:
The system segments nodes into two distinct categories: primary-tier nodes connected via low-latency high-bandwidth networks for performance-critical operations, and secondary-tier nodes connected via higher-latency lower-bandwidth networks for quorum participation. This segmentation allows even-numbered primary nodes to maintain simplicity while adding odd-numbered secondary nodes to enable quorum achievement.
Solution Approach 2:
Secondary-tier nodes act as intermediaries that participate in leader elections and quorum decisions without requiring high-speed network connections. These mediator nodes enable the system to achieve quorum with an even number of primary nodes, preventing split-brain scenarios while maintaining the simplicity of even-numbered primary node deployment.
2Quantity of substance
If nodes are placed in less secure environments to reduce operational costs, then cost is improved, but data security and integrity may be compromised
Solution Approach 1:
The system applies different quality requirements to different node tiers: primary-tier nodes in secure data centers require high security and low-latency connections for data storage and performance operations, while secondary-tier nodes in less secure environments only require participation in leader elections and quorum decisions, accepting higher latency and lower bandwidth.
Solution Approach 2:
The system creates an asymmetric architecture where primary-tier and secondary-tier nodes have different roles, security requirements, and network specifications. This asymmetry allows cost reduction through secondary nodes in less secure environments while maintaining data security through the asymmetric separation of critical data operations from quorum participation.
3Adaptability or versatility
If heterogeneous network connections are used for different nodes, then operational flexibility and cost efficiency are improved, but system complexity increases
Solution Approach 1:
The consensus ring protocol implements a universal mechanism that handles both low-latency primary-tier node communications and higher-latency secondary-tier node communications through the same protocol framework. This multi-functionality allows heterogeneous network connections to be managed uniformly, reducing the perceived system complexity despite operational flexibility.
Data Source
AI summary
Disclosed examples cause transmission of an advertisement from a first node in a first availability zone to a second node in a second availability zone and a third node in a third availability zone, the advertisement to specify a leader-candidate status for the first node, the first node and the second node eligible to vote for a leader and eligible to serve as a leader node, the third node eligible to vote for the leader and ineligible to serve as the leader node; access votes from the second node in the second availability zone and the third node in the third availability zone; and after the votes satisfy a quorum to elect the first node as the leader node, set a role of the first node as the leader node.