SAN & NAS Appliance Single VMDK or Multiple?

[maggit13]

I am setting up a 2-node HA system with SAN & NAS Free on 2 ESXi hosts (i know this is legacy). I only have 1 RAID controller on each and the HDD's are configured in RAID 5 presented to the hosts as local datastores. This means I have to use virtual disks as the storage disks on the appliances.

My question is, should I create 1 large virtual disk (say 1TB) or several smaller virtual disks (5 x 200GB)? Once inside the Web GUI, when creating a storage pool I then choose which Software RAID settings to use, either RAID0 for the single disk, or RAID5, 10, etc. for multiple disks.

I want to avoid any performance problems and I'm wondering if doing a software RAID on top of multiple vmdks is not a good idea, i.e. the single, large disk with RAID0 in the storage pool is the better way to go instead of RAID5 on 5 vmdks.

I've been looking all over for an example of virtual disk configs for the appliance and can't find anything anywhere. Any help is appreciated!

[yaroslav (staff)]

Hi,

You can go with one VMDK/RDM unless you just plan to play around with MDADM inside CVM.

Once inside the Web GUI, when creating a storage pool I then choose which Software RAID settings to use, either RAID0 for the single disk, or RAID5, 10, etc. for multiple disks.

Keep it simple, just use one array. Splitting a physical array using VMDKs and shaping them back into the MDADM will eat up a lot of performance and make the system flaky.

As a side note, RAID5 out of HDDs is not supported due to write penalty. This can result in HA devices going out of sync due to storage hiccups. See the recommended RAID configurations here https://knowledgebase.starwindsoftware. ... ssd-disks/
If you plan that RAID only for StarWind HA devices, you can use RDM. RDM eliminates the VMFS+VMDK overheads and allows skipping this fix https://knowledgebase.starwindsoftware. ... ontroller/.

P.S. StarWind SAN and NAS is a legacy product that did not allow the creation of HA devices. It has been reshaped in StarWind CVM which combines HA and NFS sharing capabilities. They have similar UI, but those are different products.

[maggit13]

Thanks for the info. So you're saying I should just create one large virtual disk in vmware, then create a storage pool with RAID0 in the appliance? Don't do multiple vmdks?

If you plan that RAID only for StarWind HA devices, you can use RDM. RDM eliminates the VMFS+VMDK overheads and allows skipping this fix https://knowledgebase.starwindsoftware. ... ontroller/.

Can I do this if I only have 1 RAID controller and have to create a datastore there for the vSAN appliance?

[yaroslav (staff)]

You are welcome!
You do not need to create a RAID0 storage pool inside the VM: it will be recognized as available storage without exposing it to MDADM.

So you're saying I should just create one large virtual disk in vmware, then create a storage pool with RAID0 in the appliance? Don't do multiple vmdks?

RDM is not pass-through. RDM allows exposing the RAID virtual disk, while pass-through implies exposing the entire controller to the VM.
For RDM, VMFS volume has to be removed from the array. You could also partition RAID in ILO or iDRAC and connect that partition of the RAID to the VM as RDM.

[maggit13]

Ok thanks, I understand now.

I meant create a RAID0 storage pool in the starwind appliance settings in the GUI, not in the VM settings. I simply had 5 virtual hard drives attached to the vm and created a RAID from that inside Starwind.

[yaroslav (staff)]

Hi,

I am glad that we saved your system from the overhead
Have a great weekend.

[maggit13]

Thank you again for the help. Now I'm having a different problem.

I have 2 vSAN nodes configured in HA, one on each Esxi host. Anytime I reboot either one of the appliances, the host name resets back to the default name, Starwind-xxxxxx and the iscsi data ip disappears from the config in the web gui, the management and replication ip's both stay to what I set them to but the replication ip goes to "Unassigned". I checked the hostname files and ip files on the nodes and they have the correct information.

Any idea why this is happening and how I can get them to stay persistent after reboot?

[yaroslav (staff)]

Hi,

The IPs should stay there.
Do you have the IP set to static? Also, do you use SR-IOV?

[maggit13]

Pretty sure everything is to static but I'll check.

Not sure what SR-IOV is but I'll check that too. I do have a PCI 10gb nic in one of the hosts that the iscsi runs on. If it's enabled should it be disabled?

UPDATE:
SR-IOV is disabled on the pci nic

[yaroslav (staff)]

Please share the screenshot of the adapter settings.
Could you please share the logs of that VM? Please restart the VM (i.e., reproduce the issue) so that it is easier to investigate it.
Once the logs are collected, please try re-adding that adapter to the VM.
Thanks!

[maggit13]

Ok, here you go. Same thing, hostname changed back to default, Data nic was cleared out.
Before reboot:

After reboot:
I changed the hostname back from here before logging in.


And here are some logs. The zip was too big to attach here so I had to remove some logs. Let me know if this is enough to see what's happening.

[maggit13]

more screenshots and logs

[maggit13]

more logs

[yaroslav (staff)]

Greetings,

Thank you I will look into them in good order.

[maggit13]

Great! thank you.