REFS and raid stripe size

[mkretzer]

Hello,

I understand that it is OK to use 4k allocation block size with REFS and W2016. But what about raid stripe size for such a repository? As big as possible?

Markus

[dellock6]

Hi,
you mean the underlying block size of the raid volume? If so, 64k is usually a good choice, as it could be also the allocation block size of the ReFS volume. 4K is the default but we have some discussions in the forums (like this one for example: veeam-backup-replication-f2/2016-refs-a ... ?hilit=64k) discussing about the choice between 4k and 64k.

[mkretzer]

Strange. I remember reading somewhere here that the current suggestion would be to use 4k as 64k is the "legacy" option and even MS sees no real benefits in using 64k anymore. But right now i do not find it.
Since the smallest allocation is 64k for the RAID i will use this might be the best option anyway.

One question: Will block clone work better with 4k blocks? Isn't it much more likely that an entire 4k block does not change in an backup and thus can be cloned?

[dellock6]

What you are trying to explain is a concept belonging to a deduplication appliance, while on ReFS things work differently.
First, those are the block size used to store files in the file system, but then Veeam writes its own blocks at different sizes, depending on the deduplication level; "local" is the default and is 1MB before compression. Suppose this block is compressed to 512KB, on a 64k refs volume it will consume 8 blocks, while on a 4k volume it will consume 128 blocks. But what is going to be cloned is the Veeam block, not the single volume block, so in both cases the final effect in terms of cloning will be the same. But in no situation refs blocks are compared to others, as there's no deduplication in refs volumes.