Patrick Varley reminded me 04 is the machine where we had the faster drive installed see CBPS-680
We can see that in the number of items in the disk queue being significantly lower and fragmentation being significantly high
disk queue for the 4 nodes:
fragmentation for the 4 nodes:
So yes Sharath Sulochana we need to perform compaction at the end of the load phase. This will ensure we have the same behaviour / performance for all nodes in the cluster.