Simplest! METABOT CONTROLLER API (where mailbox is a simple file)

API between MC and Mailbox

 Metabot Controller to Mailbox:
      - read from mailbox (as much at one time as possible)
        (Assuming that, for simplest case, MC is executed after apps on
         compute nodes are finished; otherwise, mailbox may need to
         signal the MC that enough data is available to begin metabots--
         "enough" may depend on the application)

 MC functionality:
      - (partially) parse mailbox notifications to know where to launch
        metabots; may have info about type of operation metabots should
        perform

API between MC and metabots

 Metabot Controller to Metabots:
      - launch metabot
         params: notification from mailbox (give location of data, possibly
         format info), operation to perform?, output format? stored where

 Metabots to MC:
      - notify MC when finished with operation, or phase of operation?
      - notify MC if there's a problem with data?

Issues: How is MC told what kind of tasks need to be performed?

Should the MC be aware if a metabot fails?

How to know where to launch metabots

 How to send metabot/location specific information to a large # of metabots
 efficiently

When to launch metabots

There may be more than one phase of operations that metabots need to perform, these phases may conflict if some metabots finish phase 1 before others and try to begin phase 2

Possibly 2 types of metadata, depending on apps and metabot operations:

general metadata to indicate location, size of data,timestep (chunking metabot);
more app specific metadata, like Kniss' app may want to index by parameter
  settings like tissue types, and uncertainty/risk?
(fdtree's metadata is also  general to create/remove large directory/file
  structures-- needs locations, # of levels, sizes)

[In implementation, keep read or get mailbox notifications and parse notifica- tions separate and decoupled in case of changes to mailbox]

How does MC recognize potential scheduling conflicts