what is needed for a hyperconverged HA system

[franconovik]

I want to build a hyperconverged HA system

Please can you tell me what is needed in terms of HW and then how to configure with proxpmox.

I suppose we need 3 servers each having at least 3 network nics.

Do I need shared storage ? if yes what kind of storage SAN or NAS ? How to connect each server to the storage ?

What kind of fielsystem to use: ceph or zfs ?

I mean is there a guide that can help to setup the hyperconverged system. I really do no know from where to start.

 

[bbgeek17]

Hi @franconovik ,
You need 3 servers. 10G+ NICs.
You don't need SAN or NAS. These are examples of external Central storage.
You said you wanted hyper-converged - this would be Distributed storage.
ZFS is not a suitable storage for your request.
Ceph is an example of distributed storage that is well integrated into PVE to run as hyper-converged.

You can start here:
https://pve.proxmox.com/wiki/Deploy_Hyper-Converged_Ceph_Cluster

Note, the choices you make with regards to CPU, Memory, Disks, Network will affect the outcome of your experiment.

Good luck

---

Blockbridge : Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/proxmox

 

[andrew-transparent]

Well, the first place to start is getting familiar with the official documentation. These forums already have some great posts about what you're asking - so I'd pour a big cup of coffee and dig in.

As far as hyper-convereged, that implies you will not have separate shared storage (ie NAS/SAN), and each node will have some direct attached storage. That being said, lower cost shared storage is handy for things like templates and ISO's etc. The path of least resistance in my experience has been to use Ceph since it has built in support and the implementation with the GUI is really well done.

Yes there are lots of guides, but I would start with the documentation so you understand the underlying fundamentals first and come up with a idea of what will work for you're particular setup. Its not one-sized-fits-all.

With 3 servers (which is a bare minimum for Ceph) you have the option of going the mesh route (all interconnected, no switch), or going with a switch which has the bonus of better scalability if you wanted to add more nodes. Minimum of 25Gbit seems to be the current recommendation, although you can get away with 10Gbit for testing/lab environments (I have a 3-node ceph cluster running on 1Gbit - it works, but very light workload).

PS - Please reserve some budget for your backup solution (ie PBS) - PVE and PBS go together like peanut butter and jelly. I wouldn't think of deploying one without the other, so might as well plan for it from the get-go.

 

[louie1961]

Do you really want hyperconverged infrastructure or just high availability? Ceph is actually a form of disaggregated storage more than it is a form of hyperconverged storage.