Investigate writing prefix balancer

[keith-turner]

While working on #347 I ran into an issue where a prefix within a table was not balanced evenly across the cluster. Since a common way of writing fluo applications may be to partition different types of data using a row prefix, it would be nice to able to esnsure prefixes within a table are evenly spread.

In the case of #347 each level within in the tree has the same prefix and I would like each level of the tree evenly spread across the cluster.

I am thinking it would be nice to have a an accumulo balancer that can be configured with regular expressions. The regular expression would match against the tablets end row. Every unique string matched by the regular expression would form a group id. All of the tablets within a group would evenly balanced across the cluster.

For example with tablet end rows like 01:abc, 01:af, 02:ppp, 02:xyz, 04:rat, and 04:bob and a regular expression like ^(\d\d) there would be 3 tablets groups. The groups would be 01, 02, and 04. The tablets within each group would be spread across the cluster.

I think it would be possible to create this for Accumulo 1.7.0 using ACCUMULO-2841 to configure the regular expressions for a balancer. Not sure how to configure a balancer for 1.6 ATM.

[keith-turner]

Opened ACCUMULO-3439 and submitted a patch.