Transcript Team Green IT Final Presentation - u

Information Technology in 2050
Team Green IT: Joshua Blum
and Jesika Haria
Background
• Internet use and data transfer are
ever growing
• This growth is associated with an
enormous energy consumption
• Architectural and
structural modifications are needed in
the IT infrastructure.
Issues and Constraints
• Data center inefficiencies
• Circuit inefficiencies
• Moore's Law limitations
Source: CTR, Jan 2011
Recommended Solutions
• Reduce system inefficiencies
• Integrated Si photonics
• Scale down size of Ge
photodetector to reduce power
consumption
Principles of Engineering Practice 2011 UT-MIST-MIT
International Seminar on Materials
Presentation Summary
Background (Trends vs. current technology)
• Energy consumed by IT has increased 10x since 2006 to 15% of
total energy consumption
• A PC runs at 6% efficiency, data centers run at 56% efficiency
• The IT industry accounts for 2% of the world's GHG emissions,
higher than the airline industry
• Power Usage Effectiveness (PUE), Data Center Performance
Efficiency (DCPE) and Input-Output Operations (IOPs)/ Watt are
some metrics that can be used to measure data center efficiency
By 2030, IT traffic will consume 100% of Japan's
current energy production
Energy
(TWh)
IP Routers
Internet traffic
Time
(years)
Source: METI and MIC
System Constraints
• Data centers have huge losses, mostly cooling, low utilization and UPS
• Only 2.5% of energy entering a data center goes to useful computing
Source: Rocky Mountain Institute
Circuit Constraints
• Moore's Law states that the number of transistors that can be placed
inexpensively on a circuit doubles every year
• Moore's Law is approaching its limit now
• Current metal interconnect technologies are hugely inefficient due to power
losses
P: Power
C: Capacitance
V: Voltage
f : Frequency
Source: Moore's Law, Wikipedia
Architectural and structural changes need to be made
Recommendations
Architectural Recommendations
• Nano-Data Centers
• Energy Proportional Computing
• Cloud Computing
Recommendations
Structural Recommendations
• Integrated Si photonics
• Ge photodetectors
Source: Characterizing the Business Environment
Architectural Recommendations
Nano-Data Centers (NaDa)
• P2P network
• Saves in cooling costs
• Less data travel distance
• Estimated to save 30-40% of cooling costs
Energy Proportional Computing (EPC)
•
Servers & computers typically operate at
10%-50% of utilization
Eliminates baseline power waste
Technology present in mobile device
•
•
Cloud Computing
•
•
•
Works hand and hand with NaDa and EPC
Distributes workload over many devices
Allows instant access to files
Source: Case for EPC 2007
Structural Recommendations
Si Integrated photonics
• Convergence on the "last mile" of photonics and
electronics
• Inter-chip communications & all connections less
than 5cm
Ge Photodetectors
• Used for large difference in index of
refraction from Si (4.0 vs. 3.5)
• Power consumption proportional to area
(capacitance) and applied voltage
• Need a 1000x reduction to limit consumption
curve (IT in Japan)
Source: Si Integrated Ge JFET Photodetector, J Wang
Analysis
•
•
•
Architectural modifications will eliminate system inefficiencies present
For long term energy reduction, photonics will take the lead
Goal is to reduce power consumption by 1000x by scaling down and
integrating photonic devices.
Source: Intel Corporation, 2010
Consequences
• Growth is sustainable if
technology creates efficiencies.
• As needs scale up, efficiency
must scale up proportionately.
• By making IT more efficient, other
sectors can become more
efficient as well.
Implementation
Energy (TWh)
IP Routers
Internet traffic
Time (years)
Source: METI and MIC
• In 10 years, bring in architectural changes and begin implementing
structural modifications
• In 50 years, physical limitations of photonics will be reached,
beginning of a new learning curve.
• Beyond 50 years, begin new learning curve for IT with new
technologies...
... along with time travel, teleportation, and lightsabers?