Reverse incremental Vs forever forward with synthetic fulls

[NathUlph]

I'm embarking on and veeam B&R upgrade from 8 to 9.5. We currently use reverse incrementals.

I intend to build our repository on server 2016 to take advantage of the great new features. I'm now trying to work out whether to move to forever forward incrementals & synthetic fulls.

From what I've read in v9.5 for reverse incrmentals the incremental is injected into the full backup without having to create a copy and rebuild the full, thus reducing the IO required. I can also run a synthetic full if I wanted.

Is there any reason why I should move to forever forward incrementals & synthetic fulls, what advantages are there if any.

Thanks
Nath

[foggy]

Hi Nathan, your understanding is not entirely correct. Reverse incremental is in fact the most I/O intensive one, please look at this post. Further, this post will also be helpful.

[NathUlph]

Foggy thanks for the links to the posts. They all seem to reference v8 or v9. I now understand the I/0 for each job type.

For what I understand in v9.5 using an ReFS formatted repository if you run a reverse incremental backup job the changed blocks are injected into the previous full backup & the previous blocks are ejected to an incremental backup file. Whereas in pre v9.5 an entire copy of the full backup file is created & then the new blocks injected & the previous blocks are ejected to an incremental backup file. This surely is a reduction in I/0 & backup time?

Thanks

[foggy]

Again, not entirely correct. Reverse incremental works similarly in either Veeam B&R version, but in case of ReFS repository, data blocks are not actually moved between files thanks to the block cloning functionality. It's, however, true, that with this type of repository I/O load is much less. On the other side, reverse incremental is not considered the best practice in case of ReFS due to additional fragmentation this backup mode is causing.

[NathUlph]

Thanks Foggy, much appreciated.

[pesos]

Where can we read more about this fragmentation issue? We tend to have problems crop up now and then with our synthetic full runs, and I have been unable to find any way to manually trigger/force a synthetic full... am I missing something?

We were considering switching back to reverse incremental but want to read up on fragmentation.

Thanks!

[foggy]

Hi John, nothing special to read, the VBK file gets fragmented with time due to the fact that new data blocks are merged into it during each incremental job run, while old blocks are stored in VRB files and marked as free inside VBK to be further re-used by the new ones.

That said, prior switching to reverse incremental, I'd first investigate the reasons causing issues with the synthetic full.