I'm trying to convert an existing distributed infrastructure with HPC. The existing infrastructure knows (1) the total number of compute nodes that are available, and (2) gives each one a unique number, called the
engine offset - from 0 for the first node, up to the number of compute nodes available.
Obviously this conflicts with the SOA HPC way of doing things, but:
Can I get the number of service hosts that will be used? We could then restrict the job from getting any more resources allocated and use this as our "total" number from item 1.
Can I give each service host an ID? A service host correlates to a task within a job, could I use the task id or some other method? The only other option I could think of, would be to use the common data framework from SP2 to read and write to a shared
text file to co-ordinate the ID's that are used.