Ceph » mgr » Dashboard

I prefer not to hide rotates column for now because:

• DriveGroupSpec use it to select devices (https://docs.ceph.com/docs/master/mgr/orchestrator_modules/#ceph.deployment.drive_group.DriveGroupSpec)
• A rotational device is likely to be an HDD, but a non-rotational device doesn't have to be an SSD. It can be an NVME or other device.

Kiefer Chang wrote:

I prefer not to hide rotates column for now because:

• DriveGroupSpec use it to select devices (https://docs.ceph.com/docs/master/mgr/orchestrator_modules/#ceph.deployment.drive_group.DriveGroupSpec)
• A rotational device is likely to be an HDD, but a non-rotational device doesn't have to be an SSD. It can be an NVME or other device.

Good point. As they seem related, would it make sense to combine both the type and rotational flag in one column then?

Kiefer Chang wrote:

I prefer not to hide rotates column for now because:

• DriveGroupSpec use it to select devices (https://docs.ceph.com/docs/master/mgr/orchestrator_modules/#ceph.deployment.drive_group.DriveGroupSpec)
• A rotational device is likely to be an HDD, but a non-rotational device doesn't have to be an SSD. It can be an NVME or other device.

Punch card readers would be non-rotational and are/were slower than HDDs. And, NVME is a controller/interface technology; the underlying medium is still solid-state.

I can understand that "rotational" might make sense for kernel developers regarding physical data seeking and queuing policies (https://github.com/torvalds/linux/commit/1308835ffffe6d61ad1f48c5c381c9cc47f683ec), but I always found it weird/misguiding to expose it from a user perspective. The property has_shell of Animal is good hint for their slowness metric, but Slugs or Sloths are have_shell = false and yet slow (perhaps it is the "Sl-" thing).

I'd be in favor of HDD/SDD over rotation/non-rotational for indicating IO/latency performance.