InfiniBand Labels
CKS applies several topology-related and performance-related labels to instances equipped with InfiniBand, which are useful to control the placement of workloads and monitor the network's performance. The following labels are available:
Label | Description |
---|---|
ib.coreweave.cloud/fabric | The name of the InfiniBand fabric, representing a distinct network fabric. |
ib.coreweave.cloud/leafgroup | The number identifying the leaf group within the InfiniBand topology. |
ib.coreweave.cloud/leaves.current | The IDs of the current leaf switches connected to the Node. |
ib.coreweave.cloud/leaves.expected | The IDs of the expected leaf switches that should be connected to the Node. |
ib.coreweave.cloud/ports.current | The IDs of the current ports used by the Node. |
ib.coreweave.cloud/ports.expected | The IDs of the expected ports that should be used by the Node. |
ib.coreweave.cloud/speed | The overall speed of the InfiniBand network. |
ib.coreweave.cloud/speed.current | The current speed of the InfiniBand network connected to the Node. |
ib.coreweave.cloud/speed.expected | The expected speed of the InfiniBand network for optimal performance. |
ib.coreweave.cloud/superpod | The number identifying the superpod within the InfiniBand topology. |
node.coreweave.cloud/rack | The number identifying the physical rack where the Node is located. |
When scheduling Pods, the fabric
, superpod
, and rack
labels may be used to set a Pod's affinity toward those Nodes, which is useful for some workloads and Node Pool configurations.
Learn more
GB200 NVL72 instances may also use the ds.coreweave.com/nvlink.domain
label to set a Pod's affinity toward one or more NVLink domains. See Control placement with NVLink domain for more information.
For example, this Pod affinity rule targets a specific InfiniBand fabric, superpod, and rack:
Example
affinity:nodeAffinity:requiredDuringSchedulingIgnoredDuringExecution:nodeSelectorTerms:- matchExpressions:- key: ib.coreweave.cloud/fabricoperator: Invalues:- <FABRIC_NAME>Your fabric name- key: ib.coreweave.cloud/superpodoperator: Invalues:- <SUPERPOD_NUMBER>Your superpod number- key: node.coreweave.cloud/rackoperator: Invalues:- <RACK_NUMBER>Your rack number