State of the CCS

1
State of the CCS

• SOS 8
• April 13, 2004
• James B. White III (Trey)
• trey_at_ornl.gov
• virtual Buddy Bland

2
State of the CCS

• CCS as a user facility
• CCS as a DOE Advanced Computing Research Testbed
  (ACRT)
• Future plans

3
Facilities

• Computer facility
• 40,000 ft2 over two floors
• 36 raised floor (lower floor)
• 8 MW power, 3600 tons cooling
• Office space for 450
• Classrooms and training areas
• Labs for visualization, computer science, and
  networking

4
User Facility

• CCS designated by DOE as a user facility
• Supports users from academia and industry
• Pursuing agreements with Boeing and Dow Chemical

5
User Community
Users come from all around the country

• 70 of usage is from users outside of ORNL

6
FY03 Usage by Discipline
7
CCS Usage Model

• Small number of large projects
• CCS supports liaisons for large projects
• Center can be dedicated to single task of
  national importance
• Human genome
• HFIR restart
• IPCC

8
Advanced Computing Research Testbed

• ACRT examines promising new computer
  architectures for DOE SC
• Determine usability for SC applications
• Work with vendors to improve systems
• Application-based evaluations

9
Past Evaluations
10
Current Evaluations

• Cray X1 - scalable vector
• SGI Altix - large shared memory
• IBM Federation Cluster - interconnect
• http//www.csm.ornl.gov/evaluation/

11
Cray X1

• Worlds largest X1
• 8 cabinets
• 256 MSPs, 3.2 TF
• 1 TB memory
• 32 TB local disk
• Cabinets half populated to test topology and
  facilitate upgrade

12
SGI Altix

• Large memory, single-system image
• 256 Itanium2 processors
• 1.5 GHz, 6 GF, 6 MB cache
• 1.5 TF
• 2 TB shared memory (NUMA)
• Targeting biology apps and data analysis

13
IBM Federation Cluster

• 27 p690s
• 32 1.3-GHz Power4s
• 864 total processors

• 8 p655s
• 4 1.7-GHz Power4s
• Login and GPFS

Federation vs. Colony Latency (µs) 12
19 bandwidth (MBs) 551
306 exchange(1x1)(MBs) 767 273 exchange
(32x32) 2199 394 bisection (2 nodes)
619 284 bisection (32x32) 922
321
14
Evaluation Plans

• Cray X series
• Upgrade X1 to 512 MSPs
• Upgrade to 1024 X1E MSPs
• Black Widow
• Red Storm
• 10.5 TF in 2004
• 21 TF in 2005
• Blue Gene at Argonne

• Cray XD1 (OctigaBay)
• SRC FPGA systems
• IBM Power5
• SGI Altix (larger images)
• ADIC StorNext
• Lustre

15
Questions?

• James B. White III (Trey)
• trey_at_ornl.gov
• http//www.ccs.ornl.gov/
• http//www.csm.ornl.gov/evaluation/