Skip to content

Latest commit

 

History

History
78 lines (62 loc) · 3.39 KB

README.md

File metadata and controls

78 lines (62 loc) · 3.39 KB
Raw

LARE-HPA: Co-optimizing Latency and Resource Efficiency for Kubernetes Autoscaling

LARE-HPA (Latency-Aware and Resource-Efficient Horizontal Pod Autoscaler) is an adaptive autoscaling solution designed to optimize the trade-off between latency and resource efficiency in Kubernetes. Unlike traditional HPA approaches, which rely on static thresholds, LARE-HPA dynamically adjusts thresholds and stabilization windows to handle workload volatility in real time.

Requirements

  • Ubuntu 20.04
  • Docker version 27.3.1
  • Kubernetes version 1.27.5
  • Istio version 1.21.0
  • Prometheus 2.54.0
  • kube-state-metrics v2.7.0
  • Python3 (tested on Python 3.11)

Note: The above setup has been tested with the specified versions. However, version dependencies may vary depending on the operating system and other installed programs.

Get Started

Prerequisites

LARE-HPA requires Prometheus and Istio to operate. Ensure that you select appropriate versions compatible with your cluster setup to collect metrics effectively. Once configured, deploy the target application for autoscaling, and generate sufficient load to verify that the metrics are being collected correctly.

Deployment Steps

1. Build the Docker Image

Build the container image for LARE-HPA using Docker.

Update yamls/Deployments.yaml

Modify the following fields:

  • containers.image: Replace with the local or remote path of your built LARE-HPA container image.
  • volumes.hostPath.path: Specify the local path where logs will be stored.

Update yamls/ConfigMap.yaml

Configure the Prometheus endpoint and specify the scaling target application's details. Ensure that the Prometheus server is correctly exposing metrics for LARE-HPA to use.

Deploy LARE-HPA

Run the deployment script to deploy LARE-HPA into your Kubernetes cluster.

./deploy.sh

Verify Metrics and Scaling

Ensure that the target application metrics are being collected and that autoscaling is functioning as expected.

Cleanup

To completely remove the scaler from your Kubernetes cluster, execute the following script:

./delete.sh

Citation

If you find the provided code or the corresponding paper useful, please consider citing the paper:

@InProceedings{kim2024lare,
  author    = {Kim, Donggyun and
               Kim, Hyungjun and
               Lee, Eunyoung and
               Yu, Heonchang},
  editor    = {Gaaloul, Walid and 
               Sheng, Michael and 
               Yu, Qi and 
               Yangui, Sami},
  title     = {LARE-HPA: Co-optimizing Latency and Resource Efficiency for Horizontal Pod Autoscaling in Kubernetes},
  booktitle = {International Conference on Service-Oriented Computing},
  series    = {Lecture Notes in Computer Science},
  volume    = {15405},
  pages     = {19--34},
  year      = {2024},
  publisher = {Springer Nature Singapore},
  address   = {Singapore},
  doi       = {10.1007/978-981-96-0808-9_2}
}
Kim, D., Kim, H., Lee, E., Yu, H. (2024). LARE-HPA: Co-optimizing Latency and Resource Efficiency for Horizontal Pod Autoscaling in Kubernetes. In: Gaaloul, W., Sheng, M., Yu, Q., Yangui, S. (eds) Service-Oriented Computing. ICSOC 2024. Lecture Notes in Computer Science, vol 15405. Springer, Singapore. https://doi.org/10.1007/978-981-96-0808-9_2

This paper is available at : Paper