[Gluster-users] Rebalancing and space efficiency

[Cedric Lemarchand]

Hello,

While expanding a dispersed volume 2+1 to distributed-dispersed 2 x 2+1, I constat that after migrating data, rebalance is based on files numbers, and not files sizes. For some use case it could be painful in the way files sizes could vary a lot (mine start from Go to dozen of To), which will lead in inefficient space usage on different bricks.

For the sake of argument, rebalancing a sample of files of 10x1Go and 1x21Go on a distributed-dispersed 2 x 2+1, where brick has 50Go, result in a 31% use on first 3 bricks and 11% on last 3 bricks.

1. Is there a way to tell gluster to rebalance data based on files size ?
2. Did the sharding xlator could be a solution, is it stable and usable in distributed-dispersed mode ?

Gluster version is 3.9.