SCD Computing Center Expansion

1
SCD Computing Center Expansion

• Aaron Andersen
• SCD User Forum
• 5/18/2005

2
Phase 1 SCD Internal Analysis
3
Growing Model Complexity
4
Chips Faster, Cheaper but Hotter

• Power problem
• P C V2 f
• Frequency (clock) increase driving Moores Law
• Decreasing voltage increases gate delay (hurts
  performance)
• Capacitance increases as gate count increases.
• Chip power consumption will increase 30 over 5
  years.
• Processor costs decreasing at 29/year.

5
Mesa Lab Power Limit
6
The Problem with the Mesa Lab Facility

• NCARs machine room was built during the era of
  low power/high cost computing systems (1976).
• 13000 sq ft
• 1.2 MW computer power limitation
• 80 W/sq ft cooling design
• We live in an era in which high power/low cost
  computing systems are available.
• It is not cheap to modify the power
  infrastructure of Mesa Lab Computing Facility,
  nor does it make sense without expanding the
  machine room floor.
• Site situated on bedrock.
• If we do nothing, the Mesa Lab power and design
  limitations will become a serious constraint on
  NCARs scientific computing capacity. This is
  true for both future vector and microprocessor
  systems.

7
Recap of Phase 1 Conclusions

• Three Driving Issues for a Facility Expansion
• Geoscience Exponential demand for supercomputing
• Technology CMOS is getting hotter and cheaper
• Engineering Mesa Lab facility design is
  obsolescent
• Recommended responses on three timescales
• Short term (1 year)
• Mesa Lab risk mitigation (RMH, Inc.)
• Medium term (2 years)
• Lease options (Staubach Real Estate)
• Colocation (Inflow, Inc., Level 3)
• Long term (3-5 years)
• Refurbishment of existing structures (RNL Design,
  Inc.)
• New construction (Uptime Institute, Inc.)

8
Phase 2 Analysis of Facility Options
9
Phase 2 Conclusions

• Center Green 3 was not as capable as hoped
• 26M to create a CG3 data center smaller than ML
  facility.
• No room for expansion
• Existing Computer Room Spaces on Market
• Most not adequate for anticipated power densities
  or loads.
• One that was acceptable was purchased by another
  entity.
• Strategy Moving Forward
• Act to maximize life and capabilities of Mesa Lab
  facility,
• Move forward on building a data center expansion.

10
Preservation of theMesa Lab Facility

• Perform a Risk Assessment
• Ensure capability of reliably housing
• 1 1.2 MW of computing
• Evaluate the capabilities of infrastructure
  equipment

11
Phase 3 Facility Design Process
12
Design Concepts
Strategy preserves initial capital investment as
computing hardware evolves rapidly. Majority of
cost in this type of facility in
outfitting mechanical and electrical systems.
13
Facility Design Concept
20,000 sq. ft. 4 MW
20,000 sq. ft. 4 MW
20,000 sq. ft. 4MW
20,000 sq. ft. 4 MW
20,000 sq. ft. 4 MW
20,000 sq. ft. 4 MW
20,000 sq. ft.
20,000 sq. ft.
20,000 sq. ft.
Module 2
Module 3
Initial Module
Expand laterally
14
Architectural Detail
15
Architectural View
16
Next Steps

• Conceptual Design Cost Estimate is due as I
  present this
• Developing full report on findings
• Develop funding strategies/options
• UCAR Management
• Board of Trustees
• NSF