While trying to solve the 5s proof time problem, I’ve found that fragments may be a major culprit. On one of my HDD, which always has above 50% proof time over 8s, is 7% fragmented. After defragment, its proof times are only 0.2s or less, consistently.
This is interesting, do you think?
Would be interesting to have some more samples. As the harvester does not much more than a handful of random reads once in a while, a major influence of fragmentation would surprise me.
I know the harvester only do a few random read from disks, that’s why I didn’t think of defragments until all other solutions has failed. The problematic disk is the first disk used for plotting, and I used it to test and transfer plots here and there between disks. That’s why it might have more fragments than usual. After defragmenting, it has been running a full day very smoothly, all reads are well under 0.2s.