Hardware Raid on Server 2012 R2, ReFS or NTFS?

[ATWindsor]

I have a setup with a 24 port areca controller-card, using hardware raid. (Storage Spaces from MS seem to be pretty lousy implemented, so its not an alternative as of today).

The question is, should i use ReFS or NTFS for my arrays? What are the upsides and downsides? I realize some of the features in ReFS are not as useful without storage spaces, but it still seems like a better solution for home fileserver use?

 

[Lost-Benji]

I have a setup with a 24 port areca controller-card, using hardware raid. (Storage Spaces from MS seem to be pretty lousy implemented, so its not an alternative as of today).

It's working fine for me. 90+ MB/sec writes and 600+ reads. No issues
Stop believing everything you read in forums.

The question is, should i use ReFS or NTFS for my arrays? What are the upsides and downsides? I realize some of the features in ReFS are not as useful without storage spaces, but it still seems like a better solution for home fileserver use?

Performance hit. For home server, not really needed.
Keep sensitive data that you can't loose of mirrored pairs with backup to another system. Bulk data on parity based RAID.

Having an Areca is good to hear but without actual model or detail is like saying you have a expensive make of car without what model.

 

[ATWindsor]

It's working fine for me. 90+ MB/sec writes and 600+ reads. No issues
Stop believing everything you read in forums.


Performance hit. For home server, not really needed.
Keep sensitive data that you can't loose of mirrored pairs with backup to another system. Bulk data on parity based RAID.

Having an Areca is good to hear but without actual model or detail is like saying you have a expensive make of car without what model.

Thank you for the feedback

Well, the claims are backed up by thorough benchmarks. Is 90 MB write in a "raid5"-like storage space setup?

ReFS carries a performance hit over NTFS? How big? And do you know why?

Thanks for the advise, but I already have a prefered setup for parity and backup.

Is the model that important for the choice of NTFS vs ReFS? Anyway its a 1882ix-card. with 4GB memory.

 

[Lost-Benji]

Thank you for the feedback

Well, the claims are backed up by thorough benchmarks. Is 90 MB write in a "raid5"-like storage space setup?

ReFS carries a performance hit over NTFS? How big? And do you know why?

Thanks for the advise, but I already have a prefered setup for parity and backup.

Is the model that important for the choice of NTFS vs ReFS? Anyway its a 1882ix-card. with 4GB memory.

My arrays are 8x drives in PARITY (RAID-5) per array. I have run the "powerprotected true" commands in PowerShell and benched with ATTO to confirm speeds. I also am moving media (HD movies and the likes) in all the time and watch the transfer speeds that match up.
I did run and test ReFS on two different occasions and seen 15-30% perf hit thanks to the background extra work needed. I also noted a lot more disk activity all the time.

As for the RAID card, nice card. These have balls and all I can say is, run up and array and test NTFS and ReFS against each other and see, being that you are running a HW card, you should see very little difference.

 

[Guldan]

Well it just so happens I've spent a few days testing such things.

My Underlying hardware was a Dell Perc H810 in a raid 60 (7/7/7) connected to three MD1000's. Drives are regular 4TB SATA 7200RPM. I made a 56gig Volume.

NTFS @ 64KB = 900-1200MB/S Read 500-550MB/S Write
ReFS @ 64KB = 850-950MB/S Read 300-350MB/S Write
ReFS @ 64KB + Storage Spaces Parity = 800-850MB/S Read 200MB/S Write

Handling of small files was abysmal with ReFS / SS Parity and even worse was the disk latencies ( This was transferring mostly large files over a 10gig link )

ReFS - Storage Spaces Parity- 2000+ms
NTFS - Storage Spaces Parity - 2000+ms
NTFS - Windows RAID5 ~ 500ms
NTFS - Hardware RAID5 ~ 25ms
NTFS - Hardware RAID5 + SS ~ 25ms
NTFS - 1 disk raid0 ~ 200ms

Some takeaways from my experimenting:

Hardware Raid > Software Raid
ReFS = Big hit on writes
Storage Spaces = Really big hit on writes ( in parity mode )
ReFS + Storage Spaces Parity = Massive write hit and disk thrashing

And my conclusion was I'll stick with my hardware raid60 on NTFS. I don't really see a use case for ReFS at this point but I would use Storage Spaces if I didn't have really nice hardware RAID cards and direct attached storage.

 

[jimblue]

For parity RAID, a good hardware controller blows away SS. However, 2012R2's SSD write-back cache (and tiering) seems to make a huge difference.

 

[ATWindsor]

Thanks for the info Guldan. I don't plan on using storage spaces, but ReFS seem to have some added resiliency to errors also without storage spaces. To bad the performance is so bad. I just can't understand how a company with the resources of MS can be so behind on things like file systems and raid-like software-implementations

 

[Guldan]

Thanks for the info Guldan. I don't plan on using storage spaces, but ReFS seem to have some added resiliency to errors also without storage spaces. To bad the performance is so bad. I just can't understand how a company with the resources of MS can be so behind on things like file systems and raid-like software-implementations

ReFS by itself if you don't mind sacrificing some write speed is a step in the right direction. Theoretically ReFS is far superior to NTFS when it comes to resiliency (if that is your top priority) however that said I haven't tested it enough to validate it.

Also if you are like me and have multiple file servers that are 95% full 60TB volumes, what happens when you lose a drive and it rebuilds?