New Server ReFS yes or no?

[AROCH]

Hello,

got a new backup server with 64 GB RAM, local disks and about 60 TB of space, windows server 2016.
I read a lot of bad things about ReFS, are these problems fixed now from MS and Veeam?
What are the recommended settings for ReFS? Format the partition as 64k ? What to do else?

[Skyview]

At this point I would say no. More headaches than what it's worth. Some users are testing a new fix from MS with somewhat good results, but more testing is needed still.

I would say ReFS is still beta.

[AROCH]

Thanks. I thought something like that and to be safe we will stay at NTFS for repository.
I do not want to be a beta tester

[suprnova]

Totally agree, avoid at all costs, unless you absolutely need the spaceless synthetic fulls and can risk your data. I would recommend waiting until there is a final decision on what causes the issues and if they can fix them.

[bovie2k]

For what its worth I have two Veeam Repositories both 60TB and have been loving ReFS. Before I had NTFS on the same hardware and it was so slow doing reverse incrementals under ReFS its lightning speed. I did 64K blocks since day 1 and have never experienced any issues with ReFS it's self.

[sg_sc]

Seems to me you have the specs to pull it off, but definitely go 64K if you do

[AROCH]

Thanks, but i will stay at NTFS until it can be used for productivity on large datastores without errors and crashes.

[ian0x0r]

Just add my 2 pence / cents

Been using ReFS 3.1 formatted with 64k allocation size for about 6 months now for both primary and backup copy repositories. Granted I may not have the largest workload in the world but I have not come across any issues so far and synthetic full creation times are blistering. Depending on how you design your backup target then you could have self healing capabilities to ensure data integrity as well as leveraging Veeams own storage level corruption guard (health check).

Following the 3-2-1 rule though I do also have a copy of my data on tape as well as 3 different disk locations should the absolute worst happen!

Ian

[daniel.farrelly]

NTFS on both our primary backup repos; 64k ReFS on backup copy repos. Make sure you have lots of ram.

[sg_sc]

I have tested extensively with a small setup.

Primary backup storage is a NAS with 1.75TB space presented as ISCSI storage to the Backup server (win2016) formatted as ReFS 3.1 64K.
Currently 1.51TB in use (250GB free) on disk. Total space calculated for all files and folders is 2.11TB. Some small savings thanks to synthetic fulls and ReFS fast block clone.

Secondary backup storage is a NAS with 1.99TB space presented as ISCSI storage to the Backup server (win2016) formatted as ReFS 3.1 64K.
Currently 1.74TB in use (260GB free) on disk. Total space calculated for all files and folders is 9.05TB. Huge savings thanks to the GFS synthetic fulls and ReFS fast block clone.

With savings on my backup copy job storage of +7TB in space and really fast merge times (max 5 min) on both primary backup jobs and secondary backup copy jobs, on cheap NAS with just 2 simple SATA drives in RAID1, I'm content.
The Backup server (win2016) has never used more than 1.7GB RAM for the ReFS metafile (checked using RamMap). It currently has 12GB RAM but could do fine with only 8GB I think. The jobs running are a mix of vmware VM backup, windows endpoint backup, linux endpoint backup. All the jobs get the benefits from the ReFS repository.

With ReFS you have huge benefits, but you do need some good specs to keep the backup server responsive.

PS: The savings on ReFS do not need any Dedup by Windows, it's automatically and without the need for CPU cycles thanks to synthetic fulls and fast block cloning. Windows Dedup is not possible yet on ReFS. Windows Dedup on NTFS might yield even better savings, but needs many CPU cycles to go hunting for the duplicate blocks and slows the storage down a lot.

[j.forsythe]

Hi.
I am running backups of our VMs (about 7 to 8TB) on two Repo's. One is a local StorageSpaces "RAID" with a ReFS/NTFS partition and the other one is an iSCSI target on a Synology NAS.
Both Repo's are about the same size, round about 15TB each.
I had all the luck with the ReFS Repo and it failed a couple times, I have lost 2 to 4 weeks of backups. Luckily I only used half of the partition and I was able to create a local partition with NTFS and it is working like a charm.
The Repo at the NAS is ReFS as well and it is working great to be honest, did not have one problem since I created it.
I still want to change the setting of the local RAID and replace Storage Spaces with a RAID created with the HP controller and a certain STRIP size that was recommended in the thread "REFS 4k horror story".

At the moment I would not recommend using ReFS, if you run out of luck you loose your backup data and that could cause some issues!
Maybe in the future I/we can migrate to a stable ReFS Repo and use it's benefits, like fast clone (which is really fast).

Regards,
John