Building new Windows 2016 Veeam Server

[ryanm]

I'd like to build a new physical Windows 2016 Server to use with Veeam. Would like to utilize the new dedupe capabilities inside of Windows 2016.

Just want to get an opinion as the best settings to use for the repositories. Are using NTFS repositories formatted with this command still the most current Best practice?
format <volume pathname> /A:64K /L from this post https://www.veeam.com/kb2023

These volumes will about 50TB in size once formatted.

[Gostev]

Yes, Windows Server 2016 does not change these recommendations.

[mkretzer]

@ryanm I don't know if you follow this forums posts so a quick warning: Wait for todays MS update before you enable deduplication to avoid data corruption!

[Mike Resseler]

@mkretzer

Is the KB32xxx implemented in todays MS Update cycle? And more importantly, are you aware that they fixed the issues that were in it?

[mkretzer]

@Mike Not 100 %. But i got an email from microsoft yesterday which sounded like it (at least for our REFS issues and the dedup issues were fixed in the same "problematic" update).

[Mike Resseler]

We had some talks last week with MSFT and the KB indeed solved the ReFS and Dedupe issues but it was shown (and out QC team investigated it) that it breaks something else. Especially on a VBR server with SQL it would make it unusable so that is why I want to be sure that the problem is solved

But if they add it today in the regular update cycle that means they should have solved it.

[ryanm]

I am aware of some W2016 corruption issues, but it's my understanding that only effects REFS? Was planning on running NTFS unless everyone suggests other wise.

[mkretzer]

REFS has no corruption issues that i know of, only instability. The corruption issues are with 2016 NTFS + DEDUP

[kubimike]

I run ReFS no Windup and don't have issues. I use REFS for its cloning ability. Its when you enable Windows dedup apparently the problem becomes a BIG problem.

[rendest]

There are big problems with ReFS in general. Dedup or not.

Either way, seems like Microsoft still hasn't got a clue what went wrong with the update: https://www.onmsft.com/news/microsoft-d ... day-update

[vanhovejens]

ReFS is just terrible to use when there is just a bit latency,,,
We've got quiet good performing storage which handles refs really well (minor performance to ntfs).
The slightly less performing backup storage is usable on NTFS volumes and with ReFS it's just awful,,,
guess ReFS underneath cannot handle with any io waits which results in awfull performance and it makes it just unusable.

@mike feel free contacting us we've already have an open veeam case but were rolling back to ntfs on our 'slower' boxes since we guess that a resolution isn't really in the make at this time...
The latest MS KB is installed on the repo servers only so veeam can still run and we've got the ReFS 'fixes'

[kubimike]

@rendest like what? I haven't encountered any yet.

[mkretzer]

@kubimike: Our example:
veeam-backup-replication-f2/refs-4k-hor ... 40629.html

[Mike Resseler]

@Jens,

So you have those latest fixes and it is still giving you issues with ReFS? Do you have 4K or 64K?

[rendest]

(works in same company as Jens)

4K was a complete disaster (see the horror story thread), I wouldn't recommend this to anyone, and I can't imagine why this would be the DEFAULT option.

64K improved A LOT. But that's probably due to the decrease in IO.

Memory leak + crash in the refs code in general was fixed in the KB of Jan 27th, however it didn't help with the filesystem time-outs.

The last KB which had the SQL leak was supposed to help with performance (we installed it since we have seperate repository servers which do not perform SQL queries). Long story short: performance is still poor when experiencing latency (as in, 100ms hiccup, results in a 20sec filesystem time-out where no IO is processed and Veeam is just "me zijn vingers aant draaien").

[Mike Resseler]

Stef,

Thanks for letting us know. Do you have a support call with us and/or MSFT on this?

(For all non-Flemish speaking readers... "me zijn vingers aant draaien" is difficult to translate directly but kind of means: Veeam is doing nothing and waiting until there is IO is again )

[vanhovejens]

Hi Mike, we have an open Veeam case 02060809 for these issues but contacting MS Support is terrible as mkretzer can confirm so we haven't tried yet,,, (feedback is just not to use ReFS en stick with NTFS.... why release it and promote it)
The Veeam case was created to know when we could install the 'horrible' KB on our repo servers so we can continue ReFS performance troubleshooting --> now were just waiting for any progress either from MS or Veeam

""Veeam support engineer:
Concerning the ReFS performance - I am afraid I do not have any information about that either. We are waiting for any update too, as we have many cases related to performance on ReFS repositories no solid solution yet unfortunately.
We have also some information about abnormally large fragmentation on ReFS volumes used with Veeam and it might be related to the issue, but no solid information yet.
Also according to Gostev sometimes storage spaces (in S2D) configured improve performance and merge reliability for ReFS repositories, however it is not the configuration I would advise as a first step.
""