Balanced Leader Distribution Algorithm in Kubernetes Clusters

Area of Science:

• Computer Science
• Distributed Systems
• Cloud Computing

Background:

• Containerization simplifies application deployment, with Kubernetes orchestrating containerized applications.
• Stateful applications in Kubernetes require persistent storage and strong data consistency among replicas.
• Leader-based consistency mechanisms can cause performance bottlenecks when leaders concentrate on specific nodes.

Purpose of the Study:

• To address the node bottleneck issue caused by concentrated application leaders in Kubernetes.
• To propose and validate a new leader election algorithm for improved load distribution.
• To enhance the scalability and availability of stateful applications in Kubernetes.

Main Methods:

• Developed a novel leader election algorithm designed for even distribution of application leaders across Kubernetes nodes.
• Conducted experimental evaluations to compare the proposed algorithm against Kubernetes' default leader election mechanism.
• Measured performance metrics to assess bottleneck mitigation and overall system efficiency.

Main Results:

• The proposed leader election algorithm effectively distributes leaders, preventing node concentration and subsequent bottlenecks.
• Experimental results demonstrate the correctness and superior effectiveness of the new algorithm compared to the default.
• Improved load balancing leads to enhanced scalability and availability for stateful applications.

Conclusions:

• The developed leader election algorithm successfully resolves the leader concentration bottleneck in Kubernetes.
• This approach offers a more robust and efficient solution for managing stateful applications.
• The findings contribute to optimizing Kubernetes performance for demanding, stateful workloads.