Dynatune: Dynamic Tuning of Raft Election Parameters Using Network Measurement
Fuente:
arXiv
Guardado en:
| Autores principales: | , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2025
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866909697221591040 |
|---|---|
| author | Shiozaki, Kohya Nakamura, Junya |
| author_facet | Shiozaki, Kohya Nakamura, Junya |
| contents | Raft is a leader-based consensus algorithm that implements State Machine Replication (SMR), which replicates the service state across multiple servers to enhance fault tolerance. In Raft, the servers play one of three roles: leader, follower, or candidate. The leader receives client requests, determines the processing order, and replicates them to the followers. When the leader fails, the service must elect a new leader to continue processing requests, during which the service experiences an out-of-service (OTS) time. The OTS time is directly influenced by election parameters, such as heartbeat interval and election timeout. However, traditional approaches, such as Raft, often struggle to effectively tune these parameters, particularly under fluctuating network conditions, leading to increased OTS time and reduced service responsiveness. To address this, we propose Dynatune, a mechanism that dynamically adjusts Raft's election parameters based on network metrics such as round-trip time and packet loss rates measured via heartbeats. By adapting to changing network environments, Dynatune significantly reduces the leader failure detection and OTS time without altering Raft's core mechanisms or introducing additional communication overheads. Experimental results demonstrate that Dynatune reduces the leader failure detection and OTS times by 80% and 45%, respectively, compared with Raft, while maintaining high availability even under dynamic network conditions. These findings confirm that Dynatune effectively enhances the performance and reliability of SMR services in various network scenarios. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_15154 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Dynatune: Dynamic Tuning of Raft Election Parameters Using Network Measurement Shiozaki, Kohya Nakamura, Junya Distributed, Parallel, and Cluster Computing Raft is a leader-based consensus algorithm that implements State Machine Replication (SMR), which replicates the service state across multiple servers to enhance fault tolerance. In Raft, the servers play one of three roles: leader, follower, or candidate. The leader receives client requests, determines the processing order, and replicates them to the followers. When the leader fails, the service must elect a new leader to continue processing requests, during which the service experiences an out-of-service (OTS) time. The OTS time is directly influenced by election parameters, such as heartbeat interval and election timeout. However, traditional approaches, such as Raft, often struggle to effectively tune these parameters, particularly under fluctuating network conditions, leading to increased OTS time and reduced service responsiveness. To address this, we propose Dynatune, a mechanism that dynamically adjusts Raft's election parameters based on network metrics such as round-trip time and packet loss rates measured via heartbeats. By adapting to changing network environments, Dynatune significantly reduces the leader failure detection and OTS time without altering Raft's core mechanisms or introducing additional communication overheads. Experimental results demonstrate that Dynatune reduces the leader failure detection and OTS times by 80% and 45%, respectively, compared with Raft, while maintaining high availability even under dynamic network conditions. These findings confirm that Dynatune effectively enhances the performance and reliability of SMR services in various network scenarios. |
| title | Dynatune: Dynamic Tuning of Raft Election Parameters Using Network Measurement |
| topic | Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2507.15154 |