ReFS

[WBadminUser]

So.......

Whats your guys opinion on using this instead of NTFS?

 

[Christopher (Drashna)]

I'd rather use ReFS over NTFS. I'm mostly waiting on Alex to reverse engineer the file system and create a parser, to optimize CloudDrive, and to implement proper support into StableBit DrivePool and StableBit Scanner. 

 

However, I may just convert my pool over now, and be done with it. 

 

You can read a lot about it on Wikipedia, and it's a good read. If it a bit technical.

https://en.wikipedia.org/wiki/ReFS

 

 

However, the main improvements over NTFS are:

• Maximum volume size of 1YB (that's a Trillion TBs!!!!) rather than 16TB (for 4kb allocation unit size) to 256TB (for 64kb allocation unit size), and a max file size of 16EB (millions of TBs).  
  These may not affect most of "us", but some of it may, especially with CloudDrive. 
• COW (Copy or allocate on write).  Instead of modifying data, it writes it to a new section on the disk, even when modifying data.  This is much better for power failure, other other issues.  This may be a bit more data intensive for CloudDrive, bit may experience less integrity issues.
• Integrity streams.  It stores a checksum of the data on the disk (in addition to what CloudDrive does).  
• Resilience, it should detect and correct damage when it occurs, and doesn't require taking the disk offline for maintenance. Eg, no need to run CHKDSK, as all of that is built into the file system
• Compatible with most existing Win32 disk API. 

 

The downsides are: 

• Its new enough that there are very few recovery tools. If the drive fails, recovery may not be possible or may be VERY expensive.  This will change with time, as ReFS increases adoption.  Additionally, if you're using duplication with StableBit DrivePool, the issue is at least partially mitigated
• Does not support Alternate Data Streams (at least on Windows 8, Windows 8.1 and 10 support these), 8.3 file names, file compression, encrypted files, hard links, quotas, or deduplication.
• There is a performance hit, in a large part from the Integrity streams (as these are computed on all writes).  However, the performance impact may not be significantly (less than a 10% hit), from what I've seen.

[lee1978]

I have been using refs for well 6 months and not had any issues what so ever and no corruption in fact I have to say my system has never been this stable

[Christopher (Drashna)]

I have been using refs for well 6 months and not had any issues what so ever and no corruption in fact I have to say my system has never been this stable

Added fuel to the flame of me converting.  I've been wanting to do so and have been putting it off for a while..... in part because my pool is massive and that's 22 drives to juggle data around on...

[lee1978]

Yeah I had the same a very time consuming process I had 2 new drives just in case I had a failure so I formatted them as refs added them to the pool then used the duplicate later option to remove a few drives at a time and rinse and repeat until I was done.

 

If Alex ever gets time to try implement the self heal ability of refs drivepool will be very sweet deal

[Christopher (Drashna)]

Yeah, it's a very long process.....

 

I did it a while ago to convert to 64k cluster (allocation unit size) for all of my drives (except for the SSDs). 

 

The up side is that ReFS uses this size by default (doesn't give you an option, I believe) 

 

 

And the self healing thing is something that we've talked about internally. Basically, it would be checking the integrity streams, and if it comes back with issues, check for duplicates, and replace with good copies.