[Info-vax] Defrag utility needed on Integrity SSD-only configurations?

[Hein RMS van den Heuvel]

On Tuesday, September 10, 2019 at 3:30:14 AM UTC-4, Marc Van Dyck wrote:
> Sometimes, for no special reason, the defragment script of a disk fails
> to run or to reschedule. And we clearly notice a degradation of
> performance each time this happens, and this disappears as soon as
> the defragmentation index of the disk comes back to something
> reasonable.
> I'm not able to explain why. This is purely empirical observation.

The original main intent for defrag was to reduce disk head movement - seek time. Obviously, for storage with good caches and/or SSD this is no longer relevant.
However, there is also simply a lot more CPU 'work' involved managing many little disk fragments, across many extension headers, some of it serialized behind cluster wide locks making everything suffer, not just the process needing to map one more extent.

You should be able to see and quantify this degradation with various T4 (monitor) counters notably for the file-system, lock manager, and possibly XFC. In the end you'll be using more Interupt, MPsync and Kernel time per unit of business work (or simply in comparison to the User+Exec mode time).

The (excessive?) logging you indicate may well play a big role. 
Maybe just make the volume extent size pretty big (more than 5000 blocks?)
if there are not too many very small files (such that the space waste is manageable) combine this perhaps with a very large cluster size (500+ ?) 

hth,
Hein.