Disk setup for proxmox 5 install

[robb]

I am completely new to proxmox and would like some advice on how to implement proxmox on my home server.

Hardware:
Supermicro mITX MoBo with Intel Xeon D1521 processor, 16GB DDR4 reg ECC.
Disks:
1 x 60GB SSD
1 x 250GB HDD
2 x 2TB HDD

Currently I have these disks in an old(er) HP N40L Gen7 Microserver. They will be re-used in the new hardware with the SuperMicro MoBo. Bitfenix Prodigy case.

The idea is to have NethServer 7 as a VM that will be doing the majority of tasks on my LAN, like networking services (DHCP, DNS), gateway, webserver, accounting and filesharing (Samba4 domain) etc...
Other resources on the server will be used for testing purposes.

Currently the disks are used as follows:
60 GB SSD: /
250GB HDD: /var/log/
2 x 2TB HDD's: MDADM Raid 1 /home

What disk setup and fs would you advice to use? The case can accomodate more disks and in the future it is likely I will add more 2 TB disks.

 

[guletz]

hello robb,

I want to say some facts, maybe other can say more or better:

- regarding nethserver, it is not so wise to put all the eggs in the same basket
- use different VM for different service, like dhcp, dns, samba4 and so on

For the storage I would make like this:

-250 Gb hdd for proxmox only (lvm)
- 2 x Tb with zfs mirror/aka raid1 (after the proxmox install) for any VM usage
- 1x 60 gb SSD for zfs cache(using only 32 Gb)


Also better will be to reuse your old server.


With zfs you can add more disks if you want.

 

[robb]

Hi guletz,
Thnx for the response.
I was more or less thinking in the same direction. Thanks for confirming my initial hunch.

However I am curious what you mean with: 'Also better will be to reuse your old server'.
Since this HP microserver only has 4GB of RAM and the processor is an AMD Turion series processor, I currently am having performance issues. The processor just can' t cope with the services provided by NethServer. That' s why I want to switch to the new hardware and in that same process put in a virtualization layer for more flexibility.

 

[guletz]

You can use your old server for backups made on the new one. For a friend I do like this. Proxmox on both(new and old). One time in a week, the old server is starting (with a smart outlet power, who can be scheduled), and after the backup is done, the old server is power off. If my friend want, obviously can power on the old server when he need / want(including from any remote location if he have Internet access)

 

[Rhinox]

Either add more RAM, or forget about ZFS. With just 16GB RAM (and 2TB disks!) you do not have enough memory for ARC and VMs...

 

[guletz]

Either add more RAM, or forget about ZFS. With just 16GB RAM (and 2TB disks!) you do not have enough memory for ARC and VMs..

NethServer 7 will not use more than 8GB RAM. And in the worst case, zfs can do a good job even with only 1-2 GB RAM. I have a similar server(only 12 GB) compared with NethServer, and a lot of other stuff(squid, ospf, nfs, samba, etc) with zfs mirror 2x2TB and in worst case I have at least 1-2 GB free. This server is working pretty good for me

On other server, 16 GB RAM with 4 x 4 TB(raidz2) + 2x2 TB(mirror), 2 lxc instances(mysql cluster, webmail) also works good, and 1 GB Ram free all the time.

Generally speaking, is not so importat how much memory do you have on a server. Important is if your software can working properly for your needs, with this RAM.

 

[Rhinox]

Yes, zfs can live with 1-2GB arc if your disks were small and your disk reads/writes negligible. But for effectively caching 2TB drives, you need at least 4-8GB only for arc (and some ssd for slog/l2arc). With that zfs could do much better job. Try it one day, and you'll be surprised...

If all you need is your software to work, ok, but I'm always trying to get the best for my bucks. In this case (with just 1-2GB for disk-cache) lvm+ext4 gives better results than zfs.

 

[guletz]

Yes, zfs can live with 1-2GB arc if your disks were small and your disk reads/writes negligible.

This is not a general rule. Depends a lot of what kind of load do you have. A simple example:

- if I run a rsync backup(milions of files), the only thing that is important is if the rsync metadata can fit in the arc memory, but the speed of rsync will not be helped in any other cases


ARC memory size is usefull if this data /metadatathat is allready there is usefull for others I/O operation. So if I put any data in ARC, but I do not need to used again in maximum 5 seconds (when the data are written to disks) is not important if I have 1 GB or 10 GB alocated for zfs.
This fact doesn't mean that I can have only a small I/O activity on zfs, as you sugested. In my examples, my network is the bottleneck (around 850-920 mbits). So in conclusion your post is true only 8n some particularly cases. It is not a general rule as you claim.

A big ARC size is useful IF I wrote some data now and in max 5 seconds I can use this data again for write/read. In any other cases a big size for arc is not help you as you suggested.
Anybody who use squid who use the same cache mecanics can easily understand what I am try to say (in a bad english - sorry for this)



For this reason, I react with this post. And belive me, that a I use zfs with different hardware/zfs settings starting from zfs-fuse.