Comparing Moab metascheduling to Condor and MCP (Modified Critical Path)

Project Details

Project Lead
Jenett Tillotson 
Project Manager
Jenett Tillotson 
Institution
Indiana University, Research Technologies  
Discipline
Computer Science (401) 

Abstract

One of the key components of any grid system is a job scheduler that chooses proper resources based on the feature requirements of the jobs in the queues. FutureGrid gives us a unique opportunity to compare Moab metascheduling to other possible solutions in order to make informed decisions about the future of grid computing with respect to XD.

Intellectual Merit

Using a metascheduler in environments like the TeraGrid introduces many challenges due to the separate nature of the administration of the machines but metascheduling could allow a system like the TeraGrid to be more grid-like in terms of use.

Broader Impacts

This research will impact the TeraGrid and the XD systems.

Scale of Use

I'm going to be starting with 28 VMs - 8 computer nodes plus a head node will comprise a compute cluster and I will have three such clusters. I then need 1 VM to be the metascheduler. These can be nimbus VMs just fine - meaning they don't have to be persistent. I also would like external internet access to the head nodes and the metascheduler node, but I could be fine with external access to just the metascheduler. After initial testing, I will probably want to scale it up as large as I can, but that may not be necessary. I expect this research will take several months to accomplish.