raid choise for Windows repository

[1dna]

Helloo Panel,

I hope some of you mind sharing our input.

I have a alletra 4140 with 30x20 TB on each 932i controller.

Ive started out with 30x20 in raid60 due to most amount of speeds and safety with disk in mind ( note we have another alletra offsite running raid6 and hardend repository) so we have multiple copies.

we rencently switched to this as primary (physical proxy taking backup from storage snapshots).

we are running it with 256gb of mem and 2x intel xeon gold 6526Y.

Currently we have formated it

Disk raid

Stripesize : 256KB
Raid Type: 60
Disk groups: 2

Filesystem: Refs
64kb blocks

OS Windows 2025

performamnce

1:
We have tried running it in a SOBR - but as using data locallity it will only fill on 1 controller at the time which bring down to 1 controller not being used. (is switching to performance instad a valid way?)

We have split the sobr into 2 separate repository aswell and tried splitting the jobs to raise the paralism.
Speeds varies about 1,5 GB/S to 2,5 GB/S - Reads is fast.


Asking:

Im not really sure if this is what to expact from a Repository - as Raid60 has some penalities and as we are using syntethic fulls it seems like these raid 60 is taking a write penalities. also for none cached writes.

As we have split the raids i have an option to change raid configuration as we only have moved some data and its not showing satisfying results compared to testing, but hey always something in the wonderfull world of it.

to find a balance where write speeds is prefered but data security and stability is a must - anyone can recommend a different raid level? should we go to 128 KB instead?

i hope for some input

[pybfr]

My most urgent recommendation wouldn't be to change the RAID but the file system and use Veeam Hardened Repository iso which would give a much more stable, trouble free and immutable repository, especially if you are already using in other places. There nothing a windows repository will do better than an XFS one, and a lot it will do worse.

[edh]

We largelly used RAID 60, how many colums did you used for create your raid 60, 4? 8?

As columms i wanna say raid-subsets of RAID 6, and how many disk per subset.

More subsets = more speed, but less capacity.

Modify: Sometimes more subsets is better move to RAID 10. Then you will get some extra write capacity.

[1dna]

My most urgent recommendation wouldn't be to change the RAID but the file system and use Veeam Hardened Repository iso which would give a much more stable, trouble free and immutable repository, especially if you are already using in other places. There nothing a windows repository will do better than an XFS one, and a lot it will do worse.

we have had that and have 1 offsite (for copy jobs) but our sadly Best experince is some features in veeam Works the Best on Windows,

We have 2 alletras on prem for backup - so im still considering keeping 1 linux and windows for the other stuff.

Thing is, im a sucker for data security and the best option out there - to me that would be raid 6 / 60 - but rebuilding a 30disks 20 tb raid 6 would.. not be pretty i guess.

[1dna]

We largelly used RAID 60, how many colums did you used for create your raid 60, 4? 8?

As columms i wanna say raid-subsets of RAID 6, and how many disk per subset.

More subsets = more speed, but less capacity.

Modify: Sometimes more subsets is better move to RAID 10. Then you will get some extra write capacity.

You mean groups - 2 of 15 to maintain some sort of space as we are quite heavy in data need and we have alot of backups...so i would be able to move to raid10 as it wouldent simply be enough space for us.

As i see it - only real option here would be raid 6.. and maybe do more raidset? but its a... terrible route and costs alot of disks - i get it about the data security.. but it has to balaance -we can live some of the time it takes (writes) but restore options is far superooier.

what do you run the raids on?

[pybfr]

There is no difference in term of Veeam features between a windows and hardened repository except XFS offers immutability and ReFS does not.

Plus ReFS and windows have an history of instability and problematic upgrades, there is absolutely no reason to use it today, not when hardened repo has become such a streamlined solution.

With 30 disks I would build a 3 way RAID 60 with 3*10 disks… and would absolutely avoid RAID 6 beyond 10 or 12 disks…

[1dna]

@pybfr

Well, would that increase the write speed a bit to handle the the writes better

i dont disagree - team wise windows is somehow making it easier for some - but thats another story.

can i ask - as this is the same server (2 raids never the less to 2 raid controllers) would it make sense at all to use a scale-out ? and not just use 2 seperate repositorys and move on ? and just lower the amount of tasks

all alletras we have has 2cpu and 256gb of mem.

[pybfr]

If it were my choice I would do a single large repository (without knowing any specific constraint you may have)

[1dna]

I dont disagree, but - the alletra comes with 2 seperate storage controllers with 30 disks on each seperate raid - but it also has some beneficials.

Im just trying to shake the curve before moving too much data to it, moving 100s tb of data and not +500.

we have around total of 70 Backup jobs and 70 copy jobs and tape and so on, so im trying to figure out how we can maximise this s to the fullst in the easiest way possible (Linux or Windows xfs or refs.. that is).

i cant get around it gives me 2 raids - which can become 2 repositorys 1 for each controller or a scaleout (which i believe will not maximise it ) or maybe make that much sense as the extents would still be on the "same" server

we are at the point where change in current jobs wouldent be a bigger point but changes occur.