PVE as Drop replacement for vSphere

[Imad Daou]

Hi Guys, I need your help please. I am presenting proxmox as drop replacement for vSphere here in our company. Here are the list of questions I have so far:

Questtions
###
For example the ceph file system seems to be like gluster where you have a bunch of remote shares that all act together as one. However what is the stability like? What happens if you lose 5 servers due to a power outage or a crash of some sort. Will that take out ALL of your VMs?

I am not sure as I haven’t looked at it in a while but I don’t recall KVM supporting migration. VMware can migrate a VM if its host machine dies (as they are using the same storage across the cluster) Can KVM do the same?

And how many nodes the Proxmox can hold in a Proxmox/Ceph HCI setup?
###

Your answers to those questions will be highly appreciated.

Thanks you!

Imad

 

[sb-jw]

For example the ceph file system seems to be like gluster where you have a bunch of remote shares that all act together as one.

CEPH is an Block /Object Storage, not an Filesystem. But it's now possible to add an cephFS Share, which is an RBD pool and on top of this an Filesystem. That's a native function but needed 1 or 2 clicks more in PVE GUI.
I would CEPH not compare with GlusterFS.

However what is the stability like? What happens if you lose 5 servers due to a power outage or a crash of some sort. Will that take out ALL of your VMs?

If you have an correct Crush map, crush rules, infrastructure and enough Server, you are able to lose a whole rack without any impact.
If you have 5 server and your crush rule replicate on host level, then you are able to lose one server without any impact, maybe you can lose an second one, it depends where the crush algorithm stores the data.

I am not sure as I haven’t looked at it in a while but I don’t recall KVM supporting migration. VMware can migrate a VM if its host machine dies (as they are using the same storage across the cluster) Can KVM do the same?

You can live migrate KVM too, and with HA the VM will restart on another node. But it's only HA not Fault tolerance, that's a big different!
In PVE there is no equivalent to DRS.

And how many nodes the Proxmox can hold in a Proxmox/Ceph HCI setup?

Recommend is up to 32 Nodes, maybe you can add many more but it's not recommended.
https://pve.proxmox.com/wiki/Cluster_Manager

 

[Imad Daou]

Thank you for your prompt response sb-jw,

If we decide to go with Proxmox, the cluster will be using the HCI way using Proxmox/Ceph only. No SAN or NFS will be involved. Does your answers apply to HCI as well?

I believe a 32 hypervisor nodes per Proxmox cluster cell is more than enough.

Thank you again.