S.M.A.R.T. in KVM on virtio drive

[namron]

Hi,

I use passtrough to make hdds, that are physically connected to my hypervisor available to a VM.

For example with code like this:

qm set 102 -virtio1 /dev/disk/by-id/ata-TOSHIBA_DT01ACA300_548XGR9GS

Accessing the hdd from within the VM works fine, except SMART and device info. It seems like no drive infos like vendor, serial, etc. and SMART data are passed through.

Did I miss some additional arguments in my command or is it just not possible to have it work this way?

 

[LnxBil]

Have you tried other implementations like SCSI or SCSI over virtio?

 

[namron]

No, I haven't because I thought virtio to be superior over the other alternatives

 

[LnxBil]

No, I haven't because I thought virtio to be superior over the other alternatives

It's not, e.g. SCSI over virtio-scsi does have more features, but passes everything through a scsi compatibility layer.

 

[namron]

I am mainly concerned with speed. Does the additional scsi layer impact performance?

If I give it a try which SCSI Controller Type should I use?

 

[LnxBil]

I am mainly concerned with speed. Does the additional scsi layer impact performance?

Driver are almost the same (and need to be changed on windows)

If I give it a try which SCSI Controller Type should I use?

As I already said, it's almost the same, but only through scsi, so Virtio SCSI.

 

[namron]

Allright. Changed the passthrough to SCSI and SMART works now. Thank you.

Another topic:
My host runs on an AMD A10-5700 APU. 1 socket, 4 cores.
What is an intelligent way to assign coren to 3 VMs, which experience nearly the same load?
Each VM with 1 socket, 2 cores?

 

[LnxBil]

The overhead of multiple cores is not only on the cpu, it's also on memory. In multicore systems you often need to spawn processes for each core, so your "minimal" host with only a few "real" (user land) processes have almost doubled number of processes due to a core increase from 1 to 2 cores. This yields for more used memory - it's not that much, but it is more and you also need more CPU cycles to maintain these operations and stuff. Yet in the end, I don't think that you'll have to worry about that. As long as you have a load level in your host that is lower than the number of cores you're fine. You should not assign more cores to your VM than you physically have, but I think that's clear.

In my experience, you're almost never bound by cpu power in a virtualized system, it's always I/O or memory. If you really need that much CPU power, you would not use the low-end apu for that, but a beasty dual or quad socket 24 core Xeon :-D

 

[naglerrr]

Does this really work? I am using Proxmox VE 4.4 and because of issues with PCI Passthrough (https://forum.proxmox.com/threads/p...for-passthrough-of-ibm1015.32841/#post-168599) of my SAS2008 Controller I tried using drive passthrough. I tried using sata0: and scsi0: with virtio-scsi-pci but in both cases I am unable to access SMART values inside the guest. How can one pull this off?

 

[namron]

Does this really work? I am using Proxmox VE 4.4 and because of issues with PCI Passthrough (https://forum.proxmox.com/threads/p...for-passthrough-of-ibm1015.32841/#post-168599) of my SAS2008 Controller I tried using drive passthrough. I tried using sata0: and scsi0: with virtio-scsi-pci but in both cases I am unable to access SMART values inside the guest. How can one pull this off?

It doesn't work.

I do SMART directly in proxmox now.

 

[proxwolfe]

Does it work by now? I would also be interested in checking SMART inside the VM...