Some questions about some of the new HPC 2008 functionalities
-
Friday, July 04, 2008 12:09 AMHello everyone,
I have been looking at the Simple Api and have some doubts regarding the following functionalities:
Batch Normally scheduled job.
Broker Service router job.
Service Application integration job.
- What are these types of jobs? How to use them?
- job.EndpoitAddresses (used only by service jobs)
"Retrieves the unique network addresses that a client uses to communicate with a service endpoint."
What is this for?
- job.JobTemplate
"Retrieves or sets the template to use for the service job."
What type of stuff can i define on the job templates? Is this the same as the job templates on CCS (where we would use an xml structure to define the job all the job properties?)
- node.NodeGroups / job.NodeGroups
"Defines a node group."
What are they for and what can i do with them?
- job.UnitType
"Determines whether cores, nodes, or sockets are used to allocate resources for the job."
Can anybody clarify this for me?
- And, finally, what are the CreateFilterCollection, CreateIntCollection and CreateStringCollection scheduler methods for?
Can somebody explain them to me (or point me to the correct documentation? In some cases the SDK documentation is a little vague).
Code examples would be very appreciated :o)
Thanks in advance,
Luís Silva
All Replies
-
Wednesday, July 09, 2008 7:16 PMOwnerHi Luis,
A few resources for your review:
1. See the job scheduler and WCF broker presentations at http://channel9.msdn.com/shows/the_hpc_show.
2. See the 'Getting Started Guide' and the 'Sample WCF Service' downloads at http://connect.microsoft.com, register for the 'windows hpc server 2008 beta' program.
3. See the 'resources' folder at http://windowshpc.net where there a few zip files with scheduler API sample codes.
http://channel9.msdn.com/shows/the_hpc_show -
Friday, July 25, 2008 12:51 PM
HPC v2 now suppports the HPC Basic Profile (HPCBP) of the Open Grid Forum. The Basic Profile defines a web service for execution of jobs on the grid. job.EndpointAddresses is how a submitted job is known in the HPCBP. It is its 'handle'
NodeGroups is a method of associating a subset of nodes together so that jobs run on only certain nodes in the cluster. The HPC Cluster Manager allows you to create a Node Group and add nodes to it. As an example, this could be used to run jobs on machines that had special software installed or belonged to a certain department.
node.NodeGroups is the list of groups this node belongs to.
job.NodeGroups is the list of groups that a machine must be a member of to run the job.
When you run a job, you can say that you need 5 machines (nodes) or 4 sockets (processor dies), or 10 cores to run a job. UnitType tells the scheduler which you want.
The Create*Collection methods are used to allocate the specified collections for use in other methods (such as getNodeList).- Marked As Answer by Don PatteeModerator Thursday, June 25, 2009 2:10 AM
-
Sunday, August 17, 2008 6:39 AMHow about 2048 threads? I have software that can run that many.
Developer http://contract-developer.dyndns.biz