Skip to content

Habana per-process utilization information #373

@lars-t-hansen

Description

@lars-t-hansen

Our current Habana backend does not have per-process utilization information, only per-card, see #234 for longish discussion and results of investigation. This is a regression relative to slurm-monitor, which relies on hl-smi. It would appear that hl-smi uses privileged / private APIs to get this information. We should decide how important it is for us to have it, whether it is worth the cost to run hl-smi to get it, and so on.

(Probably not a high priority at the moment.)

Metadata

Metadata

Assignees

No one assigned

    Labels

    LaterLow priority / background taskLoggingapi-missingNo API exists to get the info.slurm-monitor-parityFeature parity with slurm-monitor

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions