Advise for raid ZFS data

[fabienfs]

Hello,

I would like some advice for my data raid.

I have a Proxmox with 4 SSD 2TB for data.
(64Go ram, Intel Xeon 3.7GHz 8 core)

On this data raid I would like to put
several Windows and Linux VMs.

The but is : fast raid and the security of being able to lose 1 disc

I started by doing a raidz with default blocksize (8) and ashift 12. but I am not satisfied because it takes double the space than what I have as data.

What do you recommend ? To increase the blocksize? If so, how much? 12? and for the ashift?

Or would it be better to do a different kind of raid? Dunuin advised me this on another thread : « A striped mirror would be better for workloads like a VM storage.

Thank you for your opinions,
Fabien

 

[bofh]

several things here.

#1 raidz-1 isnt optimal at all. its like a raid5 and with such big disks not that fault tollerant as you might think. theres a chance loosing a junk.
also 4disks are suboptimal for either. problem is not total capacity but high capacity over 1.5tb a drive.

in any case you also need to calculate your volblocksize based on #of ur disks
https://docs.google.com/spreadsheet...jHv6CGVElrPqTA0w_ZY/edit?pli=1#gid=1224630924
this shows us that with 4disks volblocksize doesnt matter and you will loose a lot

#2 i might suggest raid 10 here even tough you loose 50% capcity.
reasoning is that its just 20% less efficent than raidz1 (with wonky fault tollerance) and even less vs raidz2
however it brings a lot more I/O to the table which is always a needed thing in virtualisation


even more so if you use consumer grade ssds. here we really need that i/o bad. and for 20% less efficency at least one drive each mirror is always fault tollerant, rebuilds are much faster and i/o is much better. also no "suprises" in datause because of bad allignment and padding (you get what you see, while on a multidrive raidz its always just an estimate as data occupys more space than it should because of padding)