Can i do proxmox install on slower ssd

[boxieee]

ok i have 500gb & 1tb pcie 4 980 pro nvme 6700r/5700w speed and a pm951 nvme 256gb 1200r/384w speed. (32gb ddr5 ram)

my motherboard can hold all 3 drives nvme

1) the pm951 (vz) will be 0, so what i am going to do is use 150gb for proxmox then
lvresize -l +100%FREE /dev/pve/root
resize2fs /dev/mapper/pve-root

now i have 150gb for root . more than sufficient to play with( will store backups and isos etc on 1tb nvme)

2) have 500nvme for vmz

3) 1 tb for backups and misc stuff as directory can store isos etc

My question ---- what will be the performance diff if i load the proxmox on faster 500gb nvme and leave the pm251 out. and experienced person can gudie.
because i feel as the proxmox runs it has to do read and write of some files time to time and faster drive would do that faster. or most of it is loaded in memory???

 

[bpak]

If I understand your post correctly, your want to use your "slow" pm951 for the OS, am I right?

In this case, you probably shouldn't worry, as basically the OS drive will not be called so often (except on boot), and not for the most IOPS-demanding processing.
Any Linux system can run perfectly fine even on HDD, and just as a comparison, I run my PVE on a SSD mirror (which is probably a lot slower than any of your NVMe), and never encountered any kind of issue.

 

[boxieee]

If I understand your post correctly, your want to use your "slow" pm951 for the OS, am I right?

In this case, you probably shouldn't worry, as basically the OS drive will not be called so often (except on boot), and not for the most IOPS-demanding processing.
Any Linux system can run perfectly fine even on HDD, and just as a comparison, I run my PVE on a SSD mirror (which is probably a lot slower than any of your NVMe), and never encountered any kind of issue.

If I understand your post correctly, your want to use your "slow" pm951 for the OS, am I right?

yes

 

[bpak]

So no worries, it will be fine! I guess a very large majority of the proxmox users don't have NVMe to run it, and using NVMe won't even make a visible difference in terms of usability. A SSD is far sufficient. Think that Raspberry Pi may even run on a SD card...

For the same reason, even the drive where the CT volumes reside will not be so important, as they also basically host OS installations, so not called very frequently.
In some cases, what may make a small difference though, is where your data is. But it's only relevant when accessing your data directly (like your media stuff), or when using highly demanding application (DB with huge volume of data or many users, ...).

As a side note, my 2 cents: If you care about your data, instead of dedicating your full 980 pro nvme drives to separate datastores, you may also consider to have a partition in each of them (to hold data for example), and create a Raid1 mirror with them. This will not impact performance much as you run on NVMe (better reads, slightly slower writes, who cares?), but it may improve survivability in case of hardware failure.
Just remember the sysadmin motto: Raid is not backup

 

[boxieee]

So its clear that vms are okay in this setup. What abut ct... Suppose I have plex in ct and nvidia is shared then many of the host resources will be used like drives etc right. There I feel faster drive is better again I might be over thinking it's a home server

 

[bpak]

If VMs are okay, CTs will be, as a CT has direct access to the host filesystem (and no kernel), which is not the case for a VM. In a VM, unless you use ZFS, the virtual disks are stored as files on the host (qcow2) and are managed by the VM kernel, which induces some overhead.

But I understand your question is not here, but maybe rather about the difference between an executable file and data.
What you stated in at the end of your first post ("most of it is loaded in memory") is true for executable files. Once your OS is booted, most of the executables it needs are loaded in memory (processes), and even if it may periodically launch some other executables or read/write some files (configuration, logs, ...), it is negligible in terms of disk performance. And this is approximately true for any kind of application like Plex: once it is launched, it (mostly) resides in memory.
By definition, data are not used the same way: you read/write data on-demand. For example, when you decide to watch a movie, or play a game that requires tons of textures files and so. Loading those data is more demanding in terms of disk performance, in a short period of time.
That is why, my piece of (general) advice would be: segregate your VM/CT OS volumes from your data, as the first may be less demanding than the second.

All this theory being said, coming back to your case: you will be perfectly fine.
If your concern is about watching movies with Plex, and those files are stored on your NVMe and directly accessed by Plex, there won't be any sort of issue with your disks, ever. Even HDDs would be fine, that's the typical use case of many Plex users running on Synology with a bunch of HDD.
Just take a look at the HDD vs SSD vs NVMe read/write performance, and you'll see you will be far far ahead of any HDD : 150 MB/s read for a standard HDD, ~500 for a SSD, and with your 980 Pro NVMe... 6700 ?! And we are not even talking about IOPS, where SD and NVMe will literally crush any HDD.

As a conclusion: If your use case is a home / media server, I don't see any reason to be worried: your server is going to be a beast (and many would be jealous ).
If you want to look for further disk performance optimization, you may want to take a look at ZFS, but it requires some knowledge, has some caveats, and is surely not needed in your case.