Query planner interface

[smburdick]

Since we're writing 2 different query planners/optimizers, the interface should probably be the same for both. I think the argument could just be a list of ranges, with each range containing alternating elements: machine ID's, followed by the list of vectors we want from that machine. We could put each range in its own list.

query_plan(r1 = [
m1, [v1, v2, ... vn]
m2, [v1, v2, ..., vn]
...
],
r2, r3, ...)

It could return a list of machines to visit, in a particular order. You could get away with returning just the machine IDs as the master could deduce which vectors to visit, but for simplicity each machine ID could be accompanied by the vectors you should retrieve from it.

[smburdick]

For my iter-prim I ended up using this interface

[(v0, (m00, m01)), (v1, (m10, m11), ..., (vn, (mn0, mn1))]
...

Instead, I'm giving it the vectors and options for the slaves, which is information used in the algorithm. However, at this stage it'd be worth trying it both ways, and it wouldn't be a big deal to support both if you want, and maybe pick the best one.

[smburdick]

We could write a Python function that transforms one representation to the other since they are informationally equivalent to support both formats from the C code.