Volcano

Application Scenario

The basic scheduling of Kubernetes can only allocate resources by detecting the number of Ascend processors. To implement affinity scheduling and maximize resource utilization, the network connection mode between Ascend processors needs to be identified to select the optimal network resource. Volcano, deployed on the management node, can provide network affinity scheduling for various Ascend devices and networking modes.

Component Function

• Calculate the available devices in a cluster based on the fault information and node information reported by the underlying cluster scheduling components. (self-maintain-available-card is enabled by default. If it is disabled, the information about available devices in a cluster is obtained from the underlying cluster scheduling components.)
• Obtain the expected number of resources from the Kubernetes task object, and allocate the optimal resources to the task based on the number of devices in a cluster, device type, and device networking mode.
• Reschedule a task when its resources are faulty.

Upstream and Downstream Dependencies

Figure 1 Upstream and downstream dependencies

1. Calculate cluster resource information based on the information reported by ClusterD. ClusterD is used by default.
2. Receive the task startup configuration delivered by a third-party system and select the optimal node resource based on the cluster resource information.
3. Pass the specific resource selection information to Ascend Device Plugin on compute nodes to support device mounting.