Deduplication: Replacing AVL tree with cache friendly data structure like B-tree

[vrajendra-datacore]

B-tree has better memory locality and (CPU) caching. Multiple comparisons could be done once cache line is loaded (as each node has multiple values).

Once the incoming buffer is checksum'ed, it has to be looked up into dedup_table (AVL tree) to see if calculated checksum is already present or not. The intention is to reduce the look up time with B-tree.

Any thoughts/comments if we can see any improvement in performance?