Components

• Access to compute cycles over a range of styles:

• • embarrassingly parallel (batch system crumbs)

  • dedicated MPI with high BW, low latency

  • full SMP (SGI Altix)

• Data management and data lifecycle management

• • high speed filesystem with direct cluster access (lustre)

  • networked filesystems with wide access (NFS, AFS)

  • backup and archive (tape)

• Visualization

• • large pixel displays

  • 3D projection based display

  • high speed connectivity to both storage and compute servers

• Infrastructure

• • staffing: computation experts on science teams

  • power/cooling - $$$

  • network: internal, external

Questions

• Balance: research group freedom vs. compute center's existing "menu".

• Can we improve the current funding model:

• Can we improve the staffing and responsibility model?  Currently, many slices of pie approach may limit ownership and ability to see big picture.

• Modularity is "good".  How can dependencies be reduced in an environment where staffing is short and security often trumps other concerns?