Case Study 18: Federated Trans-National Grid
Moab Workload Manager®

A.18  Case Study: Federated Trans-National Grid

Overview

   A major government organization must enable a grid which spans sites located across the country.  Both existing and new clusters will be integrated into this grid with support required for legacy resource managers and architectures.  Each participating cluster will be locally managed with its own set of priorities and policies and a tendency to favor local workload.  Participating labs may contribute multiple clusters to the grid but will retain control over when and to whom these resources will be made available.

   Much of the workload within this grid will require very large data sets and consequently, cross job migration decisions must incorporate data transfer costs and delays.  The grid must scale to over 30,000 processors, must provide full accounting, and allow grid wide monitoring.

Solution

   The solution consists of three primary parts, sovereignty, security, and inter-cluster scheduling.

Sovereignty    Moab allows complete specification of send and receive side policies enabling each cluster to specify the following:

  • which clusters, users, and groups will be allowed to access resources
  • which resources will be made available to external workload
  • when resources will be made available to external workload
  • what types of external workload will be allowed in terms of resources required and application type
  • how external workload will be mapped in terms of local accounts, queues, and quality of service
  • what resource limits, relative priority and fairshare usage targets will be applied to external workload
  • what immediate and short term network and compute resource loads will be allowed
  • what information about local resources will be provided

   With this control, if desired, a cluster administrator could custom tailor how his cluster is made available on a per requestor basis.

Security

   Moab can support mixed security models allowing X.509 based security to be used to grid connections spanning sites and simpler and more efficient key based security to be used for intra-site connections.

Inter-cluster Scheduling

   Moab can provide the full suite of cluster scheduling policies across distributed grid platforms providing significant flexibility and optimization opportunities.  Further, Moab inter-cluster scheduling policies can be fully data aware with Moab optimizing job completion time incorporating both data staging overhead and relative cluster processing speed.