crush: quantify data movement for common straw -> straw2 migration scenarios
- confirm that with equal weights no data (or almost no data) moves
- quantify how much data shuffles when weights are different (e.g., [1,1,1,2], [1,2,3], etc.)
Mostly want a
#5 Updated by Sage Weil over 4 years ago
Samuel Just wrote:
I see a significant amount of movement for 3 hosts each of which with 4 osds with weights 1 2 3 4. Attaching a tarball with the test maps and a script for doing the comparison. Update make_maps.sh to point at the right crushtool.
The mapping changes look about right to me. It'll depend on how much the weights within any given bucket vary. And good to see that mappings when weights are equal do not change at all.