RAM is consumed by TinyDB and forwarding buffers in the networking stack. We have also experimented with several other types of join queries and found similar re- sults: irrespective of the query, join-predicate selectivity and average node depth have the largest effect on query execution cost for the distributed join algorithm. For all graphs showing results for the distributed join al- gorithm, we show power utilization and result accuracy at steady state, after groups have formed and nodes are per- forming the join in-network. We do not include table dis- tribution costs in the total transmission numbers. We choose to do.
