+1 to VM split by size (i.e. files up to 1TB big) and/or (at least) per VM disk.
- I have just arrived from holidays and I've found out that my new year's full copy job failed. We use a bunch of 8TB USB3 disks in a scaleout repo to store an offsite 1st of Year full backup of some VMs. Even if I had several TBs free, one big server was unable to copy because it didn't fit (VBK file close to 7.5TB). Now I'm playing with high compression settings and we will see what happens (crossing fingers). I doubt I can (nor should) try to use spanned volumes on USB disks.
- W2016 dedup with files +1TB in size is a "non supported" scenario (or at least "not recommended", with in corporate, it's almost the same). We still use W2012R2 with VBR 9, but if we move to 9.5 it is very possible we will migrate also to 2016 to use ReFS for some jobs.
- "A single backup file is written in a single thread into the underlying storage." Our storage is able to use parallelism and works at its best with long queues. So when the job starts with many VMs, it achieves 2-3GBytes/s sustained rate. But the last 2-3 VMs (bigger than 6TB) take hours to complete (130MB/s). Our backup storage has more than 200 disks waiting for something to do (IBM XIV arrays, 7200RPM disks) but with 1 thread we are only using one disk each time.
- Multicore. Compressing/deduping files by Veeam only uses one core per job. We planned our servers/proxies with fast cores (3.4Ghz), but it would be nice to use the other 11 ones
- Windows deduplication. Big files are slow to process, to garbage collect and also stay on the same volume (no multithreading).
- Copy/move backup files from one volume to another. Sometimes it is needed. Again, one file, one thread: impossible to leverage our storage performance on parallel loads.
Now that we are in January, it is too late to ask Santa for these features. That's why I'm asking Mr. Gostev instead
EDIT: Hmm... another nice feature would be a setting to "process big VMs first". Generally, but not always, it is better to start with big VMs as soon as possible because they will be the last to finish.
Happy new year to you all.