This project is based on TensorFlow and Kubernetes, it is part of my graduation project.

This project mainly includes three parts. All this parts are about improvment of k8s scheduling mechanism for distributed deep learning workloads on TensorFlow.

Kubernetes version is 1.2.0 and TensorFlow version is 0.9.0. This project's distributed deep learning code used TensorFLow's between-graph mode.

Part 1. DLTAP method.

Deep Learning Task Allocation Priority:

Part 2. DLMAP method.

Deep Learning Mix Allocation Priority:

Part 3. NRST method.

Node Random Sampling Strategy:

Tips:

1. Some distributed tensorflow application's example codes are in this address: https://github.com/qwurey/distributed_tensorflow_demo