Running Proxmox for web application server (multiple mysql databases + php)

[bbuster]

For my company i am renting a dedicated webserver with the following specs:

• 4 x 1 vCPU for Cloud Server
• 8 x 1 GB vMem for Cloud Server
• 10 GB Shared SAN SSD storage for Datastores
• Storage for OS
• 10 x 10 GB Shared SAN SSD storage for Datastores
• Storage for Data
• 11 x 10 GB Shared Backup storage for Datastores
• Storage voor Back-up

I would like to buy a server and create a virtual machine to run the webserver myself.
What i would like to buy is:

• Barebone: 1 x HP ProLiant ML350 Gen9 8x SFF
• CPU: 2 x Intel Xeon E5-2620 v3 - Six Core - 2.40 Ghz - 85W TDP
• Memory: 1 x 32GB PC4 registered ECC RAM
• RAID controller including cables and battery: 1 x HP SmartArray P440ar 2GB FBWC SAS Controller 749796-001
• PSU: 2 x 500W HP Power Supply HSTNS-PD40 723595-101
• SSD: 2 x Samsung PM893 Enterprise SSD 1.92TB SATAIII

Would it be good to run the server with Proxmox myself?
The server i rent has enough capacity but is way more expensive than buying the server listed below.

 

[Nuke Bloodaxe]

Surely a 2-CPU Xeon setup would require 1 RAM stick per CPU? [Rest looks good.] However, that might be board dependent, but it stands out to me.

 

[bbuster]

I think the 1 x 32GB is not 1 stick but 1 selected package (so probably 2 x 16 or maybe 4 times 8)
But after checking a few more sites i see different machines which are cheaper and have better hardware:

• 2x Intel Xeon E5-2667 V3
• Memory: 128GB DDR4 (8 x 16gb)
• Raidcontroller P440ar
• CPU: 2 x 500watt

My idea to host my applications myself isn't a stupid thing? I mean, i don't have power backup or network backup but apart from that it would be okay to host it myself and have the server in my office?

 

[Nuke Bloodaxe]

That really depends on how mission-critical it is, for many people, it would be fine, but I would suggest having a UPS to allow it to shut down gracefully; you can purchase that later.
One major benefit you would gain is experience, but be careful it's not at the expense of your business
So, make sure you have a continuity plan for hardware failure etc.

Honestly, you should be fine, but be careful and do a recovery plan to follow, it'll reduce the stress if something goes wrong in the future.

 

[bbuster]

Thanks for the reply. A UPS was also on my mind but since we are renting an office i am not sure the internet is also behind a UPS. But for sure a "safe" shutdown would be better.

 

[Dunuin]

Not sure where you are from, but such a server can easily consume 40-80€ electricity per month. Then SSD wear, battery wear, that hardware is quite old and EoL without warranties, so from time to time you will have to replace stuff. Maybe a more expensive ISP contract. Don't know what you are paying for your VMs, but self-hosting isn't cheap either. Especially if you take all the endless work hours into account you are managing your server

 

[Nuke Bloodaxe]

You can, if you're feeling creative, take an old BBW battery, open it up, remove the LiPro cells, and replace them... I've done that before, it can save quite a lot of money [$14 vs $300+] Admittedly, the last one I worked on did involve snipping some legs and soldering in the legs of a new unit.

In terms of drives for units, @Dunuin is quite right. You may want to buy a new set, and they can always be moved to other units. If you have something absolutely non-critical, you can build an array with the intent for it to die/degrade, and replace units with more second-hand ones as you go.

In terms of overall cost, if doing it yourself comes in cheaper, and you end up running better hardware than what you can rent, it makes sense. I once had this in the past, where it was cheaper for me to buy a system to turn into a web server, than it was to rent the same; this was a very long time ago. These days, I run all my personal web services on Oracle Cloud, and no one could beat the price of free for 200GB storage, 4 Arm cpus and 24GB of RAM with 10TB egress data... I was paying $10 a month for a single CPU with 2GB and 50GB of storage before that.

Hmm... *thinks a moment* have you considered the above?

 

[promoxer]

2nd hand UPS are very worth buck. The only part that probably need replacement would be the Sealed Lead Acid (SLA) batteries which are very recyclable and also cheap. I put the router and switch on UPS as well, so the whole setup stays powered if my mains trip.

I don't use my machine to host sites though, it is a development/staging server for our developers, but is powered 24/7. It serves as my main workstation as well, since I would need a machine to work on anyway. If you are concerned about power expenses, there is a guy on youtube running a 25W server with the full works.

I would pay for the hosting of my production sites though, mainly to keep the hardware uptime pressure off myself.

Also, most hosting don't help much with managing the software unless it is one of those cheaper shared hosting (where they do certain things for you only because you don't have the permission to). For a proper dedicated setup, managing still usually falls on the self. Either you DIY or pay someone to do it.

 

[bbuster]

Thanks for the replies.
The server i am "renting" now is $400+ each month.
It has an SLA but it still is a lot of money for my "few" customers now that work on the web application.

I am renting an office and electricity prices are included.
Also i need full management over the server (installing whatever i need on ubuntu server).
At home i have an APC SRT3000RMXLI UPS but since i am in a rented office, i don't know if keeping the server alive also keeps the internet alive. I don't exactly know how the routers and switches are installed in the building and if they are behind a UPS (and still work if power goes down).

On the other hand, i think i need to accept those small risks considering the price difference.

Would i be able to see if hardware components start failing? (I think if one PSU goes bad i will be able to monitor it, but also when a CPU or Ram goes bad? Or even the motherboard?)

 

[Nuke Bloodaxe]

In terms of ECC memory failure, the proxmox log will start recording the ECC corrections, and the stick/slot which is being affected; I had this happen.
I've not encountered a faulty CPU report in Proxmox, but that goes into "mystery freeze/crash" territory when it happens. Diagnosing it is notoriously difficult; as one of my fellow technical flatmates can attest.
Motherboards will give you some feedback, but in terms of IC component failure, otherwise anything else will simply produce instability. Again, other than IC components, daughterboards etc I've not seen a motherboard report problems in the Proxmox logs.

Generally, tho, you're looking at syslog recording Kernel output for any spotted faults, so it's not really proxmox specific. I'm not sure if there is an inbuilt log parser that could flag keywords, and send the log entries to you via email as an alert; although, I'm pretty sure it could be done via Prometheus and Alert Manager.

So, we need more people to chime in with what they have seen

 

[bbuster]

Okay, understandable.
I think i will just need to try it and see how reliable it is.
If i am correct, i can automatically backup virtual machines right? And maybe save them to a synology?

If something happens i need to have the backups to get everything up again as fast as possible but i think this could be done? (Or do i need a different server with the Proxmox Backup Server?

 

[promoxer]

$400/mth sounds like a lot, i mean, depending on traffic and load, you might be able to get away with $100/mth or even a year. Machines are really quite powerful these days. Or spend on a CDN if most of the content are static.

The pressures of production self hosting is tremendous, especially when your customers are hounding at you... and you are the only one in the world who can get it fixed.

 

[Nuke Bloodaxe]

Your choice, I run an old server to act as a PBS unit, but I also backup locally and sftp the backups to another server, which then copies them along with everything else + changes only, to an encrypted HDD; daily units, and weekly offsite units.

So, in terms of restoration, the proxmox server itself has its local copies for quick restore, and the PBS is for bulk data backups, but slow restore.
The other drives are for ransomware attacks and the building burning down; or me spinning up the backups on a home server, and letting everything/everyone connect over Zerotier.

 

[bbuster]

I also understand that, but i previously rented a server from another company and that was awful, a lot of downtime and when i contacted them they created a helpdesk ticket which could took days to fix.

For developing we now use virtual machines on the Synology and that worked flawlessly for 2 years (which is not 1/10th as powerful as the server i am looking at). I need full control over the server and when i want that i think prices go up fast.

I am just not sure if i should buy a second hand server or a newly configured pc with AMD cpu and a lot of ram for example

 

[Nuke Bloodaxe]

I actually took the second option for a home unit, but it's rapidly gotten quite expensive as I bring it up to scratch. If I were doing it again, I would go secondhand. In a business sense, though, difficult. A decent second-hand enterprise-class machine will probably cause less grief if it is looked after.

 

[bbuster]

Okay, i think i will try to go for the secondhand E5-2667.
Thank you for your help!

 

[Dunuin]

You might also want to consider a small HA cluster, to prevent downtime in case of a hardware failure (like the failing mainbord/cpu you mentioned).

 

[bbuster]

I was also thinking about that. Maybe buy a second server as a "fallback" server, but how would i do that. Is it possible to have a second (maybe lower spec server) which starts all the virtual machines when the main server goes down? And do i need a third server to check if the first one is down or how can i do that?

 

[remark]

I was also thinking about that. Maybe buy a second server as a "fallback" server, but how would i do that. Is it possible to have a second (maybe lower spec server) which starts all the virtual machines when the main server goes down? And do i need a third server to check if the first one is down or how can i do that?

Yes, it is possible. Read about Proxmox HA Cluster.

 

[Dunuin]

Either 2 Servers + third device as qdevice with ZFS replication or 3+ servers with ceph for a real shared storage where you don't lose a minute of data on a failure.