Multitasking Elastic Scheduling Cluster in Tensorflow

Deep learning has always been widely concerned and has a good development prospect. More and more people join in. How to use the limited training resources, do more calculations, and enable more users to join the training, has always been a concern of everyone. This paper studies the distributed training cluster and realizes its efficient training mode. On this basis, a distributed training cluster with multitasking elastic scheduling is established. Furthermore, this article designs new functional modules: job management, GPU expansion, video memory control and improvements. Finally completed the design of a multitasking elastic scheduling cluster based on tensor flow.