1. The Genome Sequencing Center's Data Center Plans Gary Stiehr garystiehr@wustl.edu

2. The task Design a data center to support sequencing over the next 5-7 years garystiehr@wustl.edu

3. The challenges What network, computing, and storage infrastructure will be required to support our efforts over the next 5-7 years? We know what data Sanger-based sequencing methods generate and how to process them. But what about the next-generation sequencing technologies from companies like 454, Solexa/Illumina, Applied Biosystems, and Helicos? How much data will they generate? How much processing will the data require? garystiehr@wustl.edu

4. The challenges garystiehr@wustl.edu TechnologyRun Size (GB) Sanger (AB 3730xl)0.01 45430 Solexa/Illumina700 AB SOLiD4000 Heliscope10,000 What about other technologies?

5. The challenges So in designing a data center, we found ourselves trying to answer questions not only about the future of network, computing, and storage technologies, but also about the very uncertain future of next-gen sequencing technologies. One thing was for sure: we would need a lot of everything. One other thing was for certain: there was not a lot of space to put everything garystiehr@wustl.edu

6. The challenges Land locked  Space limited  Dense compute and disk  Massive power and cooling needs  Massive power and cooling require lots of space garystiehr@wustl.edu

7. 4444 Forest Park Ave Initial estimates for computing and storage needs over the next three years were translated into power and cooling requirements. These requirements would have necessitated a $2.5M upgrade to building electrical and chilled water infrastructure. That is before any money was spent on the data center and further expansion of the space was not possible. garystiehr@wustl.edu

8. The task garystiehr@wustl.edu 

9. 222 S Newstead Ave The University was in the process of acquiring a building across the street from 4444 Forest Park Estimates for purchase and razing of this space came to about $1M As an added bonus, this space provided about 5 times as large a floor plan garystiehr@wustl.edu

10. Power/Cooling Requirements Of course, predicting requirements in 5-7 years in advance is difficult: New computer hardware types. New sequencer technologies. New projects. Based off of historical purchase trends adjusted for current and anticipated projects, 20 racks per year: 2/3 of racks will contain disk @ 8 kW per rack 1/3 of racks will contain CPU @ 25 kW per rack garystiehr@wustl.edu

11. Data Center Requirements Power and cool an average of around 13.7 kW per rack, with some racks up to 25 kW. Desire to last at least six years before additional space is needed. With 20 racks per year, needed to fit at least 120 racks. Each rack needs redundant power paths backed by UPS and generator. Cooling system needs some redundancy. Avoid single points of failure. garystiehr@wustl.edu

12. Cooling Options Chilled water-based. -Single closed loop water piping a single point of failure? +larger initial cost, potential initial unused capacity Refrigerant-based (e.g. Liebert XD-series). -Refrigerant piping single point of failure (not sure)? -Higher maintenance costs for numerous condensers? -Components inside the data center (but shouldn’t require much maintenance?). +smaller initial cost, scales as needed (assuming piping is pre-installed). garystiehr@wustl.edu

13. Cooling Design Designed to cool an average of 15 kW per rack (with ability to cool 25 kW in the mix). N+1 redundancy of chilled water plants and air handlers. Floor grates rather than perfs. Hot/cold aisles partitioned with plastic barriers above racks and at ends of aisles. Closed loop water piping. garystiehr@wustl.edu

14. Electrical Design Redudant paths all of the way to utility: - dual utility power feeds+2MW generator - dual transformers and associated gear - dual UPS (one battery, one flywheel) - multiple panels in RPP (giving each rack access to both UPSs). Branch circuit monitoring A platform (partial second floor) built to hold additional electrical equipment. garystiehr@wustl.edu

15. Other Design Elements Can withstand 150+ MPH winds Receiving and storage areas Building monitoring integrated with campus- wide system LEED (Leadership in Energy and Environmental Design) certified. Dual fiber paths to connect to existing infrastructure garystiehr@wustl.edu

16. Surprises Out of 16,000 square feet of available floor space (including platform), only approximately 3200 square feet of usable data center floor space. Electrical and cooling infrastructure ate most of the space. garystiehr@wustl.edu

17. Construction Ground breaking to move in less than 1 year (phase 1+2). Phased build out (due to budget/timing): Phase 1 30 racks 2 chillers 3 air handlers 1 generator Phase 2 60 racks 2 air handlers garystiehr@wustl.edu

18. Construction Phase 3 90 racks 1 chiller 2 air handler 1 generator Phase 4 120 racks 1 air handlers 1 generator garystiehr@wustl.edu

19. Other Considerations Standard racks, PDUs? Power whips to racks--anticipating outlet and PDU types (e.g., 1U vs. blades). Initially trying 30A 208V 3-pole circuits, power whips with L21-30R connectors. Blades with IEC C19/C20, Disks with C13/C14. Some systems with N+1 power supplies--how to maintain redundancy? For C13, 208V or 120V? garystiehr@wustl.edu