VSAN VMDK size recommendation

[CHS2]

Hi experts,

we had recently faced issue with 8TB vmdk which was showing many absent components and going to resync state again and again.

It was recommended to have small size vmdk to avoid such issue in future.

Can someone please let me know or point to some document for recommended vmdk size on VSAN with FTT=1 on 4 node cluster?

Though 64TB size is supported but I guess with VSAN we cant create such big drive. Please advise.

[abhilashhb]

There's a very good article by Cormac on creating large VMDKs on VSAN. Take a look at it : Sizing for large VMDKs on vSAN - CormacHogan.com

[jameseydoyle]

I think the more pertinent question would be to determine the reason for the components going absent in the first place. I cannot think of any reason why having a large VMDK would be the cause of components going absent. By having large VMDKs (i.e. throwing all your eggs in one basket), the impact of these failures can be more catastrophic - whereas multiple smaller disks may result in spreading the risk.

There is technically no reason why having a large VMDK would not be recommended, even in a 4-node cluster. Do you get any details as to the cause of the absent state of the relevant components? Were hosts disconnecting from the vSAN network or being put into maintenance mode?

[CHS2]

VMware support only suggested to not to have such large object as when rebalance/resync happens, moving such huge object to another fault domain would break compliance of the object to its storage policy.

This huge object was showing non-complaint

[CHS2]

@Abhilashhb - that's a great article but doesn't not explain the recommended size of large VMDK. We are using RAID 1 as its hybrid cluster not all flash.

It will have more components as it will use 2x space. But then what is the recommended size of vmdk so the object will not become non-compliant again during rebalance.

[jameseydoyle]

Any object that requires resyncing following the disappearance of some components will have the same issue. The only difference is that bigger objects will take longer to resync and will therefore be out of compliance for longer. Once the components have been rebuilt, the size of the VMDK has no bearing on its compliance (unless there are insufficient resources in the remaining cluster nodes to completely rebuild the object).

Choosing the RAID5 policy as described by Cormac Hogan in his blog might be the better option for you as you have 4 nodes and you can make some big space savings with this option. However, this requires All-Flash diskgroups and I'm not sure what your architecture is.

To answer your first question, there no specific recommendations around the sizes of VMDKs. It all depends on your cluster configuration as to what can be supported.

[CHS2]

We have hybrid architecture so we can't use RAID 5.