LinuxCon ContainerCon CloudOpen China

GPU， as a powerful heterogeneous computing accelerator， is now being widely deployed on the cloud. In recognizing the GPU cloud potential, many users moved their GPU based workloads onto the public cloud. To serve these workloads under the SLA, the public cloud has to meet customer’s performance expectation and achieve the security requirement, while at the same time reduce the cost for their customers to use GPU on the cloud. This requires the CSP to support the multi-tenancy on multiple or even a single GPU in their implementation deployment; Supporting multi-tenancy present brings new challenges of DevOps, Security and QoS issues for multiple users among vGPUs and requires fine-grained GPU resources partition scheme. In this presentation, we would like to introduce how Alibaba Cloud approach these issues and how we balance the flexibility, scalability and security when deploying our