Skip to content

Using RDMA in Workloads

This chapter mainly introduces how to configure and use RDMA resources in workloads.

Note

This chapter is based on the example of using RoCE network cards with SR-IOV. To facilitate RDMA testing, the configured image should be docker.io/mellanox/rping-test, and the sh command should be run to prevent the Pod from exiting abnormally during the operation. Please refer to the following content for details.

Prerequisites

UI Operations

  1. Log in to the platform UI, click Container Management -> Cluster List in the left navigation bar, find the corresponding cluster. Then, select Deployments in the left navigation bar and click Create Image.

    Create Image

  2. On the Create Deployment page, use the image docker.io/mellanox/rping-test. Set Replica to 2 to deploy a group of cross-node Pods.

  3. Fill in the Basic Information and enter the following information in the Container Configuration.

    rdma_sriov

    • Network Resource Parameters:

      The spidernet.io/mellnoxrdma in the example is an example of using RoCE network cards based on SR-IOV. The request and limit values are currently consistent, and the input value should not exceed the maximum available value.

    • Run Command: To prevent the Pod from starting and exiting abnormally, add the following run command:

      para - sh - -c - | ls -l /dev/infiniband /sys/class/net sleep 1000000

  4. After completing the information input on the Container Configuration and Service Configuration pages, go to Advanced Configuration and click to configure Container Network.

    Container Network

  5. Select the created Multus CR, turn off the fixed IP pool creation function, select the created IP Pool, and click OK to complete the creation.

    rdma_usage01

Comments