[Pgi-wg] EMI Execution Service Specification
Mark Morgan
mmm2a at virginia.edu
Thu May 5 09:19:25 CDT 2011
As promised, please find included my comments from a brief read-through of the EMI Service Specification document. I'm afraid I don't have too much time to do anything particularly cogent and so I'm just going to type in thoughts as they occur to me while I read the document.
Section 1.2: I fear a one-off solution for delegation when, as a Grid primitive, it would be so much more useful to have a technique for delegation that crossed service functionality boundaries.
Why do both the ActivityManager and ActivityInfo port-types have GetActivityStatus/GetActivityInfo operations? What is the difference between them?
Page 6/45: I like the definitions of what happens regarding data staging for various types and times of failures -- this was definitely missing in the original BES and caused some pain for Genesis II.
Section 1.4: You have a typo in the first paragraph '"server data pull" and "server data pull"' should be pull and push.
Section 1.4: Why limit the spec. to two types of delegation tokens rather than make it extensible?
Page 8/45: I don't know yet what your "activity description document" is, but if it's like JSDL, a vector of them gets big quick (we had a lot of trouble with this in Genesis II and is the reason why we eventually implemented the ParameterSweep extensions).
Page 8/45: Ditto for the response vector. My guess is that you are not returning EPRs which will help a bit (though I don't agree with the practice), but I would suggest that you consider the WS-Iterator 1.0 specification or an equivalent.
Ongoing: I continue to see "vectors" of things given or returned -- Again, I strongly encourage you consider some type of iteration context like WS-Iterator.
Section 4.5: Personally I find the operation name "NotifyService" misleading
Page 26/45: A rough, high level glance at the Activity Description Document structure leads me to believe that it is JSDL with a different name (and some extensions). Why go to all the effort to re-write JSDL when you could simply extend it? If you are going to do that, at the very least you need to motivate the decision. To me, it seems unnecessary. I understand that you want to use GLUE2, but couldn't that have been done without re-doing JSDL from scratch?
-Mark
More information about the Pgi-wg
mailing list