Proxmox Setup for SSD consumer best practices

[lixaotec]

Dear folks,

I know this matter has several discussions here, and Ive looked in most of them. However im new to it and wish your recommendations regarding SSD setup and schemes in order to have goo speed and avoid wearout soon, if u plz:
Will install fresh new proxmox 6.2

my hw setup
AMD Ryzen 5 1600 - 64 GB ram
(all consumers pieces, none enterprise as I cannot afford costs)
1x Kingston 120gb
1x Samsung 850 evo 500 GB
1x Seagate 1TB 5400rpm 2.5"
1x Samsung NVMe 1TB 970

I have a couple of linux vms, and my main data holds on almost 500GB with 2x VM Windows Server 2019 (one with SQL Server, another for dev stuff)

How should I setup it? (i used to have windows server with hyperv physycally at NVMe and stored win VMs at samsung 850 evo), Also hyperv memory swap were located at NVMe. How could I achieve such in proxmox , or at least optimize for best wearout without losing much performance.

Huge thanks.

 

[lixaotec]

regarding failure approach, I have not worries about as my env is suitable for downtime of a couple of hours, in order to rebuild it, instead of having a replication or such that i dont need.

Therefore NVMe is a piece that I do not have replacement. So I was planning to use it eventually for some operations. would like my setup not count with it all the time.

 

[upcnuwft]

I'd recommend to install Proxmox on the 120 GB SSD and use the NVMe SSD as VM storage.

You could also do the following to reduce wearout:

sysctl vm.swappiness=1

See here for more information.

And you should do regular backups of the VMs.

 

[lixaotec]

Does it affects ext4?

I´ve plan using ext4 as fs.. why should I go with ZFS?

thanks

 

[Dunuin]

ZFS is great if you want your data to be safe. EXT4 will not fix corrupted data itself, ZFS will do. ZFS creates checksums of every datablock, scans it regularily if a datablock mismatches the checksum and if you got parity (raidz, mirror, files stored multiple times on same drive) it will repair the damaged datablock using that parity data. Its basically a selfhealing filesystem and this way you don't encounter bit rot. Backups are quite useless if your files get corrupted and you don't recognize it (or maybe months/years later). If you don't keep your backups forever your newer backups will overwrite the healthy files with the corrupted files until there is no backup with healthy files left.

Raid is not all about preventing downtimes. You need it for fixing corrupted files too.

 

[lixaotec]

ZFS seems a great alternative. Just wondering my consumer SSD will suit for it without wearing out soon.

Regarding wearout does ZFS and EXT4 differs somehow? I mean, will its load be more agressive than ext4, or it has nothing to do?

 

[Dunuin]

ZFS needs a lot of RAM for caching (8GB for example) if you care about performance and ECC RAM is recommended because otherwise you can't be sure that the checksums are valid.

 

[ness1602]

No it doesnt need a lot of ram, it is only the way it is set by default. As for the ECC, it is recommended always not in ZFS case only