Scheduling WG

The working group is in the process of creating (and perhaps modifying existing) CTSS kits to support advance reservation, co-scheduling, and automatic resource selection. Some general information:

Notes on creating a CTSS kit

The Co-Scheduling Capability Kit contains the server-side components needed on a system to support co-scheduling:

The Advance Reservation Kit contains the server-side components needed on a system to support advance reservation:

Client-side tools to allow TeraGrid users to request advance reservations and co-allocations are in the Application Development and Runtime Support kit. This is described in the definition and implementation plan of that kit.

The Metascheduling Capability Kit contains the server-side components needed on a system to support metascheduling or automatic resource selection:

[edit]

Advance Reservation and Co-Scheduling

The scheduling working group examined three tools that support advance reservations and co-scheduling: GARS, GUR, and HARC. We did not feel that we could recommend that any of these tools not be deployed on TeraGrid as they each have their own strengths. Furthermore, GUR and HARC already have existing users. We therefore decided that the best course of action is to have the developers of each of these tools create CTSS capability kits for their tools at their own pace. The working group will assist in this process and ensure that the kits produced are high quality. See our Advance Reservation and Co-Scheduling Report for further details.

One important question that many resource providers and users have is what is the impact on jobs in the batch queues if a system allows users to reserve resources at fixed times? We know of the following work that addresses this question:

Other information:

[edit]

Automatic Resource Selection

The working group is examining using Condor-G, Gridway, GRMS, and MCP for automatic resource selection on TeraGrid.

[edit]

Urgent/On-Demand

SPRUCE

[edit]

Batch Queue Prediction

The scheduling working group is evaluating different approaches to providing predictions related to jobs that are or may be submitted to batch scheduler queues on TeraGrid compute resources.