Streams and Eddies

Consider a hash join being performed between a data set that is available
locally and another one that is arriving as a data stream. 
What syncronization barriers are there?
How many moments of symmetry?
Make assumptions needed about the specific algorithm used, but be consistent.

Answer the above problem if both data sources are arriving as data streams,
rather than one being available locally.

Suppose I have to compute the conjunction two selection predicates -- one
says "sex = male" and has selectivity of 50%, the second says
"salary > 100000" and has selectivity of 1%.  Suppose each predicate takes
the same amount of time to evaluate -- call that one unit.
Suppose I am evaluating this as a scan followed by predicate application.
In a traditional query plan, which predicate should I apply first?
For a relation of one million tuples, how many units will it cost?
With eddies, what will happen, and what will be the cost?
(Ignore cost of scan).

Suppose I have an index available on salary, and have determined that this is
much cheaper to use than to scan and filter.  Discuss (qualitatively) what
impact this will have on the eddy-based query plan.