- passed directly to margo_init, creates dedicated ES's to use for RPC
  handling, which in turn causes more pmdk operations to be issued
  concurrently

- set LD_LIBRARY_PATH more broadly to avoid library not found problems
- force placement to specific nodes to match target location
- concurrent job crashing for unknown reasons at this point

- fall back to using LIBRARY_PATH to populate LD_LIBRARY_PATH
  (still not clear why rpath doesn't work)
- don't build custom uuid library

- need to find correct solution later; this is a hack

- for convenience it's easier if persistent storage is on first node
  of allocation to create and destroy pools

- differentiate file names by rank to allow single node tests
- use fallocate instead of fopen/fwrite to generate file
- error checking