Abstract
Microservice applications typically employ a technique known as connection subsetting to ensure resource-efficient and stable communication. In this technique, upstream containers selectively route requests to a limited subset of downstream counterparts via persistent connections. However, the interdependency in microservice applications and complex runtime environments pose significant challenges for effective connection subsetting, rendering traditional strategies notably inefficient. In this paper, we present Microns, a connection subsetting framework designed for microservices in shared clusters. At the application level, Microns effectively handles the complex call dependencies in applications and meticulously determines the number of connections maintained by each pair of dependent microservices. At the microservice level, Microns manages the connection relationships between dependent containers according to their respective contributions on end-to-end latency. Experiments across microservice benchmarks and large-scale simulations demonstrate that Microns achieves a significant reduction on end-to-end latency by over 74.4% compared with the state-of-the-art strategies.
| Original language | English |
|---|---|
| Article number | 26 |
| Journal | Proceedings of the ACM on Measurement and Analysis of Computing Systems |
| Volume | 9 |
| Issue number | 2 |
| DOIs | |
| State | Published - 29 May 2025 |
Keywords
- Connection Subsetting
- Microservices
- Shared Clusters