I've updated the Advanced Filter specification to include a mechanism to retrieve dynamic attributes from a node: https://forge.gridforum.org/sf/docman/do/downloadDocument/projects.ogsa-hpcp... Comments welcome. Steven
Steven Newhouse wrote:
I've updated the Advanced Filter specification to include a mechanism to retrieve dynamic attributes from a node:
https://forge.gridforum.org/sf/docman/do/downloadDocument/projects.ogsa-hpcp...
Comments welcome.
Minor nits: In the "Status of this Memo" section on the front page, you've got "Advacned" instead of "Advanced". In the paragraph at the top of page 4, you've got "activies" instead of "activities". In the paragraph describing "/hpcp-af:UserName" you've got "derivied" instead of "derived". In the paragraph describing "/hpcp-af:Owner" (also pp.4) you've got "activites" instead of "activities". In the description of "/hpcp-af:CompactResources" you've got "muti" and "mulit" instead of "multi" and you probably should use "head-node" instead of "headnode". Major point: Is there a reason for not using the elements from the Usage Record v1.0 specification in there? I can understand not wanting to necessarily provide a full UR, but it'd be a good idea for at least the elements to be semantically compatible. (As a separate point, it's not exactly clear to me what the meaning of "free memory" is; operating systems do some funny tricks there.) Donal.
Hi Donal, Thanks for the typos - embarrassing!
Major point: Is there a reason for not using the elements from the Usage Record v1.0 specification in there?
I went back and reviewed the copy of the UR spec I've been carrying about for the last month! The need is to report a node status in terms of information to drive scheduling. Here used and free memory/processors are important. I could not see a way of capturing all that information from UR. UR is also focused on the absolute resource used by a job - as opposed to a node's resources. Am I missing something here?
(As a separate point, it's not exactly clear to me what the meaning of "free memory" is; operating systems do some funny tricks there.)
True - my flip answer its what the OS tells the scheduler for it to use in making its placement decisions. It may not be correct - but at least you get consistency that way! Steven
Steven Newhouse wrote:
Major point: Is there a reason for not using the elements from the Usage Record v1.0 specification in there?
I went back and reviewed the copy of the UR spec I've been carrying about for the last month! The need is to report a node status in terms of information to drive scheduling. Here used and free memory/processors are important. I could not see a way of capturing all that information from UR. UR is also focused on the absolute resource used by a job - as opposed to a node's resources.
Am I missing something here?
Oops, I must have misread things there. How embarrassing for me! In that case, I think I'll have to amend my suggestion to request that the element names make it very clear that these are "job-independent per node totals" or something like that. Don't want anyone else making the mistake I did. :-) Donal.
participants (2)
-
Donal K. Fellows -
Steven Newhouse