Migrating to PVE from overkill linux-ha cluster - HW spec / storage question

[ovis]

Hi everybody,

first off, please don't kill me, I am new to PVE in productive environments, coming from the network engineering side - physical servers are not my strong suit, but trying to advise boss on how to proceed. I just need to come up with an approximate budget and who to ask for a server overhaul and would appreciate your two cents in terms of what makes most sense.

So current lay of the land: small-business cluster server setup, virtualized database POS backend on centos, windows 2012 for AD control and SQL server

- two physical servers with elderly Xeons, 8 GByte mem each (I think), and a 2TByte hw RAID of sorts each
- each running elderly CentOS, clustered using Heartbeat, drbd, fencing through iLO, and lots of legacy scripts
- two KVM machines:
-- 1: ancient CentOS with a low-load db server (intersystems cache) used for a tiny number of POS terminals (will be updated eventually)
-- 2: Windows Server 2012 as AD primary for a (very) small domain with ~10 clients, SQL Server 2014 as backend for another industry-specific software, and a couple of other processes (will be updated eventually too)

- backup currently through SEP onto two NAS boxes, and on VM filesystem level into a S3 bucket

The VMs have very moderate CPU and mem usage and storage when I look at them (i.e. the current six year old setup is still overkill), but uptime is critical for business continuance.

Personally, not sure if a clustered approach is even beneficial for this use case (ie how much can be gained in terms of uptime), as long as server-grade hardware and hardware / storage redundancy are used.

Would you go the PVE cluster route or just setup a single server with some level of storage pool redundancy?
3 or more nodes would be required for this to make any sense, right? Is Ceph pure overkill for this?

Thanks a lot, appreciate your input.

 

[mailinglists]

Didn't read your whole post, but I felt sorry, 'cause i do not think people will want to do your work for you (employed to know this things , so here are my quick answers.

I would always set up cluster with ZFS and replication on Enterprise SSD drives with NVMe as log and / or special devices, no L2ARC.
I would not set it up as a HA cluster, but would manually migrate VMs in case of such needs. Two nodes are enough, third can be QDevice. I would however setup replication and backups. Please note that replication is not true HA, even if enabled withing PM GUI (replication of data is not real time). You cold always do a true HA setup, but then you would need shared HA storage + HA network or shared locally atached storage and good old clustered LVM.

For more info I suggest you ask for consulting services with ProxMox g.m.b.h. or whathever they are, or get an outside contractor who knows what it is doing. I would worry if such mission critical systems would be set up or designed by someone who is asking questions like you.

 

[ovis]

Wow, thanks a lot....I half expected something like this.

I tried to clarify that this is neither my area of expertise nor my job, and I am just trying to find out what these contractors you suggested are going to want to sell....

but you‘re right here’s not the place to ask these questions, and I‘m probably not the right guy to ask them, so never mind.

thanks anyway