Ternary Computing for a Human-Cyber

advertisement
NSF Workshop
2011.9.19-20
INSTITUTE OF COMPUTING
TECHNOLOGY
Ternary Computing for a
Human-Cyber-Physical Universe
Zhiwei Xu
Institute of Computing Technology
www.ict.ac.cn
zxu@ict.ac.cn
The FIT Initiative of
Chinese Academy of Sciences
• One of the seven Frontier Research Projects
• Bio, Space, Earth, Climate, Fission, Coal, IT
– Future Information Technology utilizing human-cyberphysical resources (ternary computing)
• A 10-year basic research project
– Targeting applications and markets of 2020-2030
– Addressing China’s needs in 2020-2050
• Main components:
–
–
–
–
functional sensing
customizable internet
cloud-sea computing
science of information ecosystems
China’s Needs (2020-2050)
• Change into sustainable development with the four
simultaneous, historical constrains of
– globalization, industrialization, urbanization, informatization,
The industry sector will dominate
the national economy for decades
GDP% 1st
2nd
3rd
1993
19.5
46.6
33.9
2010
10.2
46.8
43.0
2050
5.8
42.5
51.7
The urbanization
rate will increase
from
49.7% in 2010
to
80.0% in 2050
The IT market will
increase
from
$0.15T, 400M users
in 2010
to
$2T, 1.2B users
in 2050
• Need computing for the masses, ternary computing
Z. Xu and G. Li, Computing for the masses, Communications of ACM, October
2011,vol. 54, no. 10, pp.133-141.
Example: Industrialization
• >200 million migrate workers in China
• 2010 China furniture industry: $140B
• Manufacturing equipment: 50% cost of is IT
– Needs smart equipment: current 3%  40%
Smart curve saw:
25 meters/minute
0.1  0.05 mm
saved 6 KW power
Polish Machine
• Expertise-enabled Computer Numeric Control (E2CNC)
– Expertise: domain knowledge, professional experience, know-how
Example: Urbanization
• >200 million households in urban China, >4 million added every year
– Need IT to help popularize a sustainable life style
• Electricity consumption by Beijing households in 2008:
– 11.63 billion KWH, 16.7% of the total electricity consumption
– Per-household KWH: 15000 (high), 1200 (low), 600 (green), 1320 (policy)
• China’s CO2 emission (tons) in 2008:
– 5.96 billion (total), 4.5 (per capita), 2.7 (household), 0.96 (green household)
Timely acquire massive and accurate field data from
100s millions households, for each appliance (lamp,
refrigerator, etc.) in every household.
with one sensor per home
Electricity Computing: let the physical world do the job
• Grid search and behavior optimization
Example: Informatization
• 485 million netizens in China now (CNNIC, 2011.7)
IT Market
CAGR
IT Users
(Million)
IT Spending
per Capita
1. Google
2. Facebook
2000 (Actual Data)
0.026
25.0%
22.5
$21
3. YouTube
2008 (Actual Data)
0.11
12.7%
270
$85
4. Yahoo!
2050 (Poverty Line Growth)
0.25
2.0%
1,200
$190
5. Baidu
2050 (Value-Augmenting Growth)
2.0
7.1%
1,200
$1,321
6. Wikipedia
7. Blogger
8. Windows Live
An Internet C2C service (Taobao, cf. eBay)
– 2010: >200M users (80M UV), >2.5M vendors (>50% women), 2B items 9. Twitter
10. QQ
– $59B GMV (2.5% of $2.35T), 10M items delivered/day, ~$16/item
Historic Data and Projections
•
IT Market
($ Trillion)
Alexa
Top Sites
– 2014 (estimation): $300B GMV, 32B items (merchandise & services)
• Increase delivered value (or value/item) at low cost
– Human-aided big data mining & analytics (20PB  200PB)
– Big data augmented C2B
– Better platforms: 1 week  3 months data; response 2.6  1.1s
16. Taobao
17. Sina
21. eBay
1960-2000 vs. 2010-2050
• Man-machine symbiosis  Ternary Universe (The Net)
• The scope and objects of computer science are changing
– Cyber Computing  Ternary Computing
– Turing algorithmic science  algorithm Net science
– Moore’s law  Network Effects

Algorithmic
Science
New
Information
Science
Example of Utilizing Ternary Resources
200 million families’
Cyber world
Electricity
Energy Saving:
consumption
behavior
In 2009, an average household
in China consumed 1044 KWH
Human Society
(habits, economic
But a green households
in Beijing
only consumed 600 KWH
incentives,
social
relationship)
By 2030, household electricity
consumption could be reduced by
30% through sensing Bill
and promoting green practices

Automatically sense human society and physical world

Search optimal behavior of electricity consumption
Upgrading Household Appliances:
New energy-saving
Human meterappliances
reading as data-intensive as
Promote
best practices of energy consumption
bytes/month
Rolls-Royce15aircraft
engines

electrical appliances
physical behavior of
using electricity
15 GB/month
Physical World
Ternary Computing Research Is Starting
• Professional challenges
– The DARPA Red Balloon Challenge
requires integrating Human-Cyber
resources
• Major research initiatives
– EU FET Flagships proposals (e.g.,
FuturICT) involve ternary integration
• Specific research results
– ReCAPTCHA utilizes Human-Cyber
resources
– SignalGuru utilizes Human-CyberPhysical resources
Connectivity and Integration of
People, Machines, Things
Connectivity
FIT Scope
Seamless
SignalGuru
Grid Search
Connected
Red Balloon
Disjoint
ReCAPTCHA
Traditional
Turing
Computing
None
Digital
Information
Functional (behavior, cognition)
Integration
Level
Capability Upgrade through FIT Innovations
Informationization
Capability
Function sensing of physical world and human society
Evolvable internet with end-to-end quality assurance
Sea-cloud computing handling ZB scale (1021 bytes) of data
FIT
Innovations
Pervasive intelligent services with Human-Cyber-Physical
integration (billions of users, trillions of devices)
Material Device
Acquisition
Transmission
Processing
Application
Equipment
Physical parameter sensing
Information
Technology
Best-effort packet switched networks
Cloud computing handling PB scale (1015 bytes) of data
Internets of Things, Media, Services
(100s million users, billion hosts)
Informationization
System
Addressing
constraints of
power and
security
Speed, power, software complexity trends
the Three 100-million issues
Exaflops (10
18)
Datacenter for
100’s M (108) users
World Top1 computer speed (Flops)
ICT computer speed (Flops)
ICT computer system software (LOC)
ICT computer power (W)
100 M (108) LOC
100 M (108) W
Needs:
Maintain growth in
performance, but
control power &
system software
complexity
2020
Functional Sensing
•
•
•
Compressive Sensing
Eliminating redundant data in the data acquisition phase (A-to-D  A-to-I)
Functional Sensing
– From sensing physical parameters, sensing information, to sensing behavior
• Function: formalized cognition or behavior
– Learn from biological perception networks
•
Goal: further reduce sensed data amount by 1~2 orders of magnitude
Information reduction
Functional
Sensing
Cognition
information
Behavior
Customizable Internet
• Main features
– Extend endpoints to physical devices and people
– Programmability, isolation (slices), high performance
– Behavior cognition and cross-layer optimization
• 2010-2015:
– Enable research (host-oriented, content-centric, etc.)
– basic research and testbed experiments
Non-IP
IPv6
IPv4
PEARL routers in one physical network
Each PEARL routers provides 4 Gbps ports and customizable data/ctrl planes,
and support 128 virtual routers
G. Xie et al, “PEARL: A Programmable Virtual Router
Platform”, IEEE Communication, July 2011
Cloud-Sea Computing
• Sea Computing
– A new computing model
– hierarchically self-organizing
resources of front-end nodes
– to generate local intelligence
– to perform 90% sensing data
processing
– There will be many sea terminals
• Sea-Cloud Computing
– Cooperatively divide and schedule
computing tasks at the sea side and
the cloud side
– Big data processing and massive
serving are carried out in the cloud
• Optimize performance/energy ratio
Ecosystems Science
User Experience
Service
EDS, Andersen, …
Application
MS Office, SAP, …
System Software
Windows, Unix, …
IBM, Accenture, …
Facebook Tencent
MS Office, SAP, …
Apple BEA, LAMP, …
Oracle,
Google
Android, Windows, Linux…
Machines
HP, Cisco, Dell, …
HP, Cisco, Lenovo, …
Components
Intel, Seagate, …
Intel, ARM, Seagate, …
Middleware
IBM
DEC
……
Oracle, BEA, …
1955-1980
Vertical
Physical Information
Applications (IIS)
1980-2005
Horizontal
Media Information
2005-2030
End-to-end ecosystems
Social Information
Smart grid, public safety, intelligent traffic systems, etc.
Value in productivity, sustainability, welfare and well-being
Functional sensing, customizable internet, sea-cloud computing
FIT architecture, ternary computing models, security and privacy
Foundation (CCF) systems
Phenomena, metrics, laws, abstractions, mechanisms
Time/space complexity
Energy complexity; effort complexity, sensor complexity
Systems (CNS)
Open Problems
• What are the new workloads?
– “real” workloads open to academic community
• What should be the new metrics?
– Beyond Linpack and flop/s
– Can we calculate energy complexity for each application?
• What is a good stack?
– What new properties? How to evaluate a stack?
• How to deal with the “Classis Insecta Paradox”?
– Current IT: mammals (5000 species)
– Future IT: insects (5 million species)
New Systems
Architectures
• Need computing systems enabling
– personalization, specialty,
and large volume
• Learn from IBM 360 in 1964
– Computer family and computer architecture
• To deal with the “Classis Insecta Paradox”,
we propose
– Computer tribe
– Elastic processor
Q. Guo, T. Chen, Y. Chen, Z. Zhou, W Hu, Z.
Xu, Effective and Efficient Microprocessor
Design Space Exploration Using Unlabeled
Design Configurations, IJCAI 2011.
Elastic Processor
• Has user definable microarchitecture that can be changed
dynamically to adapt to applications’ requirements
• 2 orders of magnitude improvement in power efficiency
Current chip design solutions
Many US scientists are
researching similar
issues
MOPS/mW
ASIC
Hardwired
solution
FPGA
Field and gatelevel
reconfiguratio
n
NSF Expedition project:
Customizable DomainSpecific Computing
www.cdsc.ucla.edu
Elastic Processor
GPP
Software
solution
Flexibility
GreenDroid at UCSD
Utilization wall
Conservation cores
IEEE Micro, 3/4 2011
Thank you!
zxu@ict.ac.cn
Download