PACI Futures

1
PACI Futures

• On the brink of large changes that revolve around
  Grid-based Computing
• Need to serve science initiatives that are
  wanting unprecedented levels of connectivity,
  data sharing, analysis, e.g.
• BIRN Bioinformatics Imaging Research Network
• NIH-funded. Starting with 5 MRI sites, plans to
  grow to 100
• GEON Geology Network
• Large ITR proposal to connect geologists, results
  of studies into a data and analysis grid
• GriPhyN Grid Physics Network
• ITR funded to build a Tiered and Distributed
  analysis Grid for High-energy Nuclear Physics
  experiments
• CERN Large Hadron Collider Higgs Particle
  Detection (gt1PB/yr)
• LIGO Laser Interferometer Gravitational-Wave
  observatory
• Sloan Digital Sky Survey (15TB)

2
Distributed Terascale Proposal

• 45 Million hardware acquisition from NSF
• Start of a high-performance computing grid
• SDSC/NCSA/Caltech/Argonne Partnership for
  proposal due April 19.
• More than just FLOPs called out in the RFP
• Data handling
• Large-scale compute
• Visualization
• Coupled analysis
• Expanding high-performance Grid is key to success

3
The Grid Landscape Computing, Data, Sensors
                               
 
 
more
More Aggregate CPU
more
less
Cavepic
Less Coordination/uniformity/reliability
less
DTF
Applications will target different grids to solve
problems.
Loosely Coupled Grid
More Aggregate Data Storage
Internet Grid
4
Needs/Opportunities in the Community

• Commodity uProcessors (e.g. Intel, Alpha, Power,
  UltraSparc) are the building blocks of
  high-performance compute engines
• The same machines will be tuned and augmented for
  high-performance data handling/analysis
• Access to remote machines and data essential for
  scientific research information content in
  remote datasets.
• Analysis of data, both algorithmic and visual,
  key to understanding
• Straightforward (and Secure) mechanisms to gain
  access to this high-performance grid

5
Functionality Encapsulation

• High-performance scalable clusters
• Essential Grid Services Scheduling, Security,
  Single Sign-on, resource reporting
• Scalable Visualization Engines
• High-performance access and query of distributed
  data
• Human interfaces to monitor and control resources
• These give rise to natural software development
  and packaging
• Good technology transfer means that domain
  scientists who want smaller scale versions
  systems can concentrate on science (not
  system/grid administration)
• Must strive to support each component well and in
  a manner that can go forward with technology
• Standardized software and hardware essential to
  growing the DTF Grid.

6
Fabric of Computing

• Science Grids will have resources of diverse
  capability
• Some managed professionally at Centers
• Others managed by hobbyists
• A large fraction of tools need to be open-source
  for community to adopt, especially HPC-specific
  tools
• Long history of vendor tools going away
• Community is Unix-centric. Linux is becoming
  dominant.
• Data analysis capability will become equally
  important to FLOP capability
• Network pipes will be large. Multigigabit WANs
  becoming common.
• DTF assumes gt OC-192 (10 Gigabits)

7
Where Rocks Fits In

• High-performance Cluster computing fabric
• Builds on Industry Standard RedHat Linux
  Distribution
• Extensible to add new functionality (support of
  multiple hardware platforms, networks)
• Uses standard interfaces/tools wherever possible
• Simplified build so that professional
  administrators and scientific hobbyists can
  benefit.
• Essential building block of scalable, commodity
  data analysis engines and visualization systems.

8
Where NPACI fits in

• NPACI has expertise
• Grid-based Data handling (SRB)
• Visualization Walls
• Simplified interfaces (eg. Hotpage Portal)
• As a high-end resource site for academic
  computing
• Similar software/hardware architecture to other
  resources in computing fabric
• Unprecendented levels of capability for
  breakthrough science
• Clearinghouse for HPC-specific software stack.