Artificial Intelligence and the Data Center
- 1. AI and the Data Center Jose Ruiz, VP, Engineering
- 2. Hype, Fear and Where We Are • We are a long, long, long, way from SkyNet • The possibilities are legion • But don’t believe everything that you hear • Ex: Your job isn’t going away tomorrow • AI is still in its infancy • Including within the data center • Current focus is on efficiencies 2
- 3. What is AI? • Artificial Intelligence is actually an “umbrella” term • Three components: • Neural networks • Machine learning • Deep learning • The AI “stack” • Higher level functional capability as you move up the stack 3
- 4. The Stack Deep learning Machine learning Neural networks • Deep learning • Analyzes data at different abstractions • Uses multiple neural network layers • Machine learning • Learn through absorption of information • Refine through algorithms to determine “optimal” solution • Neural networks • Computer to look like a brain • Multiple nodes • Collectively can be “taught” to solve higher level problems 4
- 5. The Stack in Action • “Intelligence” builds as you move up the stack • Learnings become more global • Information gathered in neural network nodes • Higher order patterns learned at machine learning level • Complicated data is analyzed by breaking it into component parts • Long iterative process 5 Deep learning Machine learning Neural networks
- 6. Limitations • Nuances • Difficulty with inferences • Don’t recognize causal relationships • Ex: Symptoms and disease • Doesn’t have “common sense” • Couldn’t predict and solve a problem based on activities with out detailed algorithms • Would limit the availability of “off the shelf” solutions • Now, not necessarily forever 6 (Someday)
- 7. AI in the Data Center • Currently company specific • Province of the big boys: • Google, etc. • Due to long ”learning cycle” • Example: • Google has used AI to reduce energy use by 40% • Analysis of large volumes of data • Energy used per component, outside air temp, etc. • They were capturing all along • “Taught” by algorithm to analyze interplay between variables • Determined optimal relationships 7
- 8. The Future of AI in the Data Center • Expect ”productized” applications within 5 years • Will be subtle • Incorporated into existing offerings • Ex: DCIM • 80-90% intelligence “built-in” • Customer will be responsible for fine tuning • Early focus • Energy usage and efficiency • Cooling • Server optimization 8
- 9. Summary • Artificial intelligence is an umbrella term • Progressive functionality via stack • Neural networks • Machine learning • Deep learning • In its infancy for data centers • End user proprietary • “Productization” is coming • But not for awhile • Will be subtle • Aid in increasing automation 9