What is the "rank of a node"?

[devreal]

Problem

The MPI 5.0 RC 1 on page 443, Section 9.1.3 describes the three attributes that MPI_COMM_WORLD must have. One of them is "MPI_IO"

MPI_IO: Rank of a node that has regular I/O facilities (possibly myrank). Nodes in the same communicator may return different values for this parameter.

Do we mean the rank of a process with regular I/O facilities? We don't assign ranks to nodes (and don't properly define what a node is).

UPDATE:

Later on, we say:

The value returned for MPI_IO is the rank of a processor that can provide language-standard
I/O facilities. For Fortran, this means that all of the Fortran I/O operations are supported
(e.g., OPEN, REWIND, WRITE). For C, this means that all of the ISO C I/O operations are
supported (e.g., fopen, fprintf, lseek).
If every process can provide language-standard I/O, then the value MPI_ANY_SOURCE
will be returned. Otherwise, if the calling process can provide language-standard I/O,
then its rank will be returned. Otherwise, if some process can provide language-standard
I/O then the rank of one such process will be returned. The same value need not be
returned by all processes. If no process can provide language-standard I/O, then the value
MPI_PROC_NULL will be returned.

The term processor seems off here too. And we're missing the MPI-ification of "process".

Proposal

Not sure. Replace with

MPI_IO: Rank of a process in MPI_COMM_WORLD that has regular I/O facilities (possibly the rank of the current process). Processes in the same communicator may return different values for this parameter.

Impact on Implementations

None

Impact on Users

None

References and Pull Requests

[jeffhammond]

This feature is obsolete. There used to be machines that couldn't do I/O from all nodes. Those don't exist anymore, and even if they did, MPI I/O could be implemented in the same way it was on e.g. Blue Gene.

We should deprecate and delete this attribute.

[VictorEijkhout]

Networks of workstations? V.

[jeffhammond]

Describe such a system that cannot do IO from every node.

[wgropp]

I agree with Jeff. In addition, the standard never specifies what it means to do IO, so even if you are told that a node can do I/O, you don't know what this means, other than (perhaps) that printf won't crash. This should have been removed at the same time as MPI_HOST, which has exactly the same issues.

[devreal]

That makes sense. That's not going to happen for 5.0 so are we going to keep the existing text and remove it in 6 or fix the text for 5.0?

[jeffhammond]

Just deprecate in 6.0. I don't see any point in spending time fixing something nobody uses. There's zero upside for the investment.