The map required a synchronous call to the server. The ets table
allows for concurent reads without synchnoizing the reader processes.

The benchmark suite shows no speed up from. However that is not
surprising as the results from the suite do vary heavily between
runs.

The registry showever up in a observer output with a non-zero
message-q during benchmark runs. This change makes that go away.

Initialized the free IP pool with a random order and time stamp to
build a queue in the ets table. Released IPs are appended to the end
of the free queue (through the timestamp).