Highly Available Distributed Router Leader Election via Candidate Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cluster computing systems, particularly in distributed routers, face challenges in selecting a highly-available cluster leader that ensures seamless failover and minimal disruption during control-plane failures, with existing consensus algorithms like Paxos and Raft facing limitations in scalability and stability, especially when dealing with a large number of cluster leader candidates.
Innovation Solution
A novel method and software for selecting a highly-available cluster leader in distributed routers, utilizing a partially synchronous approach with timeouts and redundancy among cluster leader candidates, leveraging Element Managers (EMs) and Cluster Managers (CMs) to ensure reliable leader election and seamless migration of control-plane services without impacting outward behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing consensus algorithms like Paxos and Raft are used for cluster leader election, then leader election can be achieved, but scalability and stability deteriorate when dealing with a large number of cluster leader candidates
Solution Approach 1:
The patent segments the cluster elements into two distinct groups: voting cluster elements that participate in leader election and non-voting cluster elements that do not. This segmentation reduces the complexity of leader election by limiting the number of active voters, thereby improving scalability while maintaining election reliability through the coordinated participation of a subset of elements.
2Ease of operation
If a cluster leader is elected to host control-plane services, then centralized management is achieved, but failover disruption increases when the leader fails
Solution Approach 1:
The patent implements preliminary actions by having non-voting cluster elements continuously monitor the status of voting cluster elements and prepare to assume leadership roles. When a voting cluster element fails, the system already has identified candidates ready to take over, enabling seamless failover without disruption to control-plane services.
Solution Approach 2:
The patent provides beforehand cushioning by maintaining a redundant structure where multiple voting cluster elements are prepared in advance to serve as backup leaders. This cushioning ensures that when the primary leader fails, the system can immediately switch to a backup without service interruption, thereby improving failover availability while maintaining centralized management through the elected leader.
3Adaptability or versatility
If all cluster elements participate in leader election, then election fairness is improved, but inter-CLC communication overhead increases
Solution Approach 1:
The patent segments cluster elements into voting and non-voting groups, reducing the number of elements that need to communicate during leader election. This segmentation maintains election fairness among voting elements while significantly reducing inter-CLC communication overhead by excluding non-voting elements from the election process.
Solution Approach 2:
The patent applies partial action by having only a subset of cluster elements (voting cluster elements) participate in leader election rather than all elements. This partial participation reduces communication overhead while still providing sufficient fairness and representation among the voting elements to make a valid leadership decision.
Data Source
AI summary
A distributed routing system is provided for use in a communication network, wherein the distributed routing system includes at least one cluster comprising a first plurality of cluster elements from which a second plurality of cluster elements is selected, wherein each of the cluster elements comprised in the second plurality of cluster elements is configured to operate as a cluster leader candidate and wherein one of that second plurality of cluster elements is selected on a temporary basis to act as cluster leader.