Data Placement Issues

[Vek17]

I am having some data placement issues when running active fulls against my SoBR. When active fulls occur it appears that all or most fulls get pushed into a single extent based on Veeam's data placement policies resulting in one extent processing TBs of backups while others get left with very light incremental backups. As you can imagine having a single extent eating the majority of the backup load results is drastically worse performance and total time required in backups then I would otherwise expect.

As I understand it Veeam does an estimated storage consumption for each assigned task and uses that as the extent's freespace when doing data placement where it prefers extents with the most absolute free space. This behavior makes sense in many cases but is it possible to alter this behavior to provide a better load spread during active fulls? The only thing I have see is https://www.veeam.com/kb2282 which would make this storage calculation more accurate, but I expect I would encounter largely the same behavior.

[veremin]

Can you shed some light on your setup? What SOBR policy is in use? How many extents you have? How many extents are allowed to store full backups (in case of Performance policy)? How loaded those extents are (in terms of total used space/GB and percentage value)? Thanks!

[Vek17]

Sure, the SoBR is using a data locality policy due to the benefits of fast clone. Here is a basic dump of stats of the extents.

[veremin]

There is a comprehensive logic behind extent selection, but you're correct in saying that, other things being equal, extent with the most absolute free space is selected. Currently, there is no way to change this behaviour.

Thanks!

[LBegnaud]

Our current idea is to monitor for job tasks that involve a full, and if we see that quickly tune down the task count on the sobr extents such that the tasks start on different extents.

Can you provide some insight on a light command that could give us the current tasks and types? We can easily monitor for sessions with "Active full" in the name, but that won't catch all fulls that can possibly run.