FINAL CALL on SPMDApplication (was [Fwd: SPMDApplication draft 6 and request for parallel application names])
Hi, I have not received any comments on the draft posted last week and there are no outstanding issues. As we discussed on the last call I am putting this now in final call within the group. The plan is to submit it to the OGF Editor for public comment at the end of the week (Friday 16). For those who may not be familiar with the process, after submission to the OGF Editor this document will enter a public comment period of 60-days so you will have plenty of time to comment later on. -- Andreas Savva
Andreas Savva wrote:
I have not received any comments on the draft posted last week and there are no outstanding issues. As we discussed on the last call I am putting this now in final call within the group. The plan is to submit it to the OGF Editor for public comment at the end of the week (Friday 16).
For those who may not be familiar with the process, after submission to the OGF Editor this document will enter a public comment period of 60-days so you will have plenty of time to comment later on.
Question: Is it deliberate that there is no mention of OpenMP in the list of parallel environment names? Arguably it is "deliberate" since OpenMP apps do not require any special actions to execute (we say nothing about the building of apps in any case); a JSDL consumer could handle all that sort of thing transparently since non-OpenMP apps will be unaffected. But this might come up during Public Comment. NumberOfProcesses: I'm not sure whether type is described correctly; I don't know how to apply pseudoschema to nillable values. :-) ThreadsPerProcess: There is an analogous problem here. I'm wondering whether it is possible to use the SPMD stuff to (try to) circumvent the resource allocations from the Resources section. If it is theoretically possible to do that, we should state some Security Considerations that make it clear that JSDL consumers should take care to ensure that the allocations derived from the Resources bound any requests from the SPMD Application. That's everything I can think of. :-) Donal.
[I've talked with Donal about this already f2f.] [Apologies if you get multiple copies. I seem to have occasional problems sending email while out of the office.] Donal K. Fellows wrote:
Andreas Savva wrote:
I have not received any comments on the draft posted last week and there are no outstanding issues. As we discussed on the last call I am putting this now in final call within the group. The plan is to submit it to the OGF Editor for public comment at the end of the week (Friday 16).
For those who may not be familiar with the process, after submission to the OGF Editor this document will enter a public comment period of 60-days so you will have plenty of time to comment later on.
Question: Is it deliberate that there is no mention of OpenMP in the list of parallel environment names? Arguably it is "deliberate" since OpenMP apps do not require any special actions to execute (we say nothing about the building of apps in any case); a JSDL consumer could handle all that sort of thing transparently since non-OpenMP apps will be unaffected. But this might come up during Public Comment.
Semi-deliberate. We did not make any strong attempt to tackle OpenMP though I think some parts of the specification (number of threads) can be used to setup the necessary environment for execution. I'm tempted to just leave it for public comment.
NumberOfProcesses: I'm not sure whether type is described correctly; I don't know how to apply pseudoschema to nillable values. :-)
ThreadsPerProcess: There is an analogous problem here.
I'm not sure what 'the right way' is to describe nillable values in pseudo schema is. I think what's there is not wrong and the schema is correct, at least. I'll leave it as is.
I'm wondering whether it is possible to use the SPMD stuff to (try to) circumvent the resource allocations from the Resources section. If it is theoretically possible to do that, we should state some Security Considerations that make it clear that JSDL consumers should take care to ensure that the allocations derived from the Resources bound any requests from the SPMD Application.
There is already some text (albeit in an Example section, 5.3.7) describing the relation between Application and Resources. I'll promote it to another section, probably in the SPMDApplication element description.
That's everything I can think of. :-)
Thanks. I think there are no major issues so I will update the draft and submit to the OGF Editor early next week. -- Andreas Savva
participants (2)
-
Andreas Savva -
Donal K. Fellows