Transcript Document
Exploring the Market for Computation Contact: www.business.duq.edu/faculty/davies Computation Solutions Traditional High-Performance Computer (HPC) Pro: Node-node communication Con: 20x to 200x cost of other solutions Traditional Cluster Computer Pro: Less expensive to maintain & upgrade Con: Requires significant infrastructure Internet Grid Computer Pro: Massive power on demand Con: Less adequate for massive data Enterprise Grid Computer Pro: Harness existing infrastructure Con: Limited power Total Annual Spending on HPC $18,000 $16,000 $14,000 millions $ $12,000 $10,000 $8,000 $6,000 $4,000 $2,000 $0 1995 1996 1997 1998 1999 Purchases of New HPC 2000 2001 2002 Spending on Existing HPC 2003 Price per GF for HPC-Generated Computation Worldwide (excluding Departmental class) Installed base of HPC Annual cost of installed base Average annual cost per GF 400,000 GF $16.2 billion $41,000 Cluster Costs 20-Node Cluster Computer Purchase price per node $1,100 Effective life 3 years Weight (including racks) 1 ton Power consumption 8 kilowatts Required air conditioning 1 ton Required space 16 square feet Computational power 12 GF Price per GF for Cluster-Generated Computation Annual Cost of a Cluster (per GF) Total = $2,300 per GF per year $533 $533 / year for labor $763 / year for nodes (amortized purchase price) $763 $125 / year for installation and configuration (amortized) $125 $33 / year for space $33 $466 / year for electricity $466 $275$275 / year for additional hardware (amortized) $69 $69 / year for software (amortized) $69 /$69 year for hardware service contracts NodescostHardware Software Service Contract Electricity Annual of a cluster > 125%Hardware of the purchase price of the nodes Space Installation & Configuration Labor Price Comparison Annual Cost per GF Traditional HPC Traditional Cluster Internet Grid Enterprise Grid +Assumes *Cost ½ availability at $100 per year. is zero at the margin. $41,000 $2,300 $300+ $0* Cost Effectiveness Cost Effective Solutions Cost effectiveness of a particular solution depends on: • Amount of work to be completed • Value of the completed work Cost Effectiveness $60,000 Cost per GF $50,000 $40,000 Computation not cost effective. $30,000 $20,000 $10,000 Computation cost effective. $0 $0 $50 $100 $150 $200 $250 $300 $350 Thousands Minimum Value of Completed Work Work = 10 GFY Work = 20 GFY Work = 50 GFY Cost per GF Cost Effectiveness and Competitive Niches $60,000 • Grids $50,000 • Cluster computers $40,000 • Traditional HPC $30,000 • Grids • Cluster computers $20,000 $10,000 • Grids $0 $0 $50 $100 $150 $200 $250 $300 $350 Thousands Minimum Value of Completed Work Work = 10 GFY Work = 20 GFY Work = 50 GFY Distribution of Computation Users 50% of computation users need less than 13 GF of power (equivalent of 20 top-of-the-line PC’s). Computation Users By comparison, the Top 500 list of most powerful computers starts at 400 GF (equivalent of almost 700 PC’s). 50% of users 1 10 100 Power (GF) 1000 10000 Market for Computation Estimated market for computation: $35 billion Computation Users Estimated market @ under 13 GF: $5 billion 50% of users 1 10 100 Power (GF) 1000 10000 Emerging Applications • UK insurance companies to help detect fraud via voicebased lie detection • Homeland Security monitors hundreds of thousands of critical locations via web cam “event recognition” • US Air minimizes aircraft/crew downtime via “traveling salesman” problem • Scottish distillery manages arrangement of barrels in inventory via in-depth search • EmPower Partners (Philadelphia) determines socially/economically optimal entrepreneurial niches in urban areas via deep data analysis Center for Data Analysis Duquesne University’s Center for Data Analysis Duquesne University is seeking partners to build a Center for Data Analysis. The CDA will bring together cutting-edge researchers in statistical analysis and massive distributed computing power. Objective: Conduct deep data mining of terrorism data, financial data, clinical data, etc. using an Internet grid computer. Our goal is that the CDA become financially self-sustaining. Exploring the Market for Computation Contact: www.business.duq.edu/faculty/davies