power management: a battery’s perspectivescipm.cs.vt.edu/slides/3.ci-song.pdf · power...
TRANSCRIPT
Power Management: A Battery’s Perspective
Dr. Song CiAssistant Professor of Computer EngineeringIntelligent Ubiquitous Computing LaboratoryUniversity of Nebraska-LincolnApril, 9th, 2009 at NSF SciPM Workshop
Motivation
• Status Quo– Current research on power management
focuses on power-aware hardware/software design and development
– However, in most cases, a battery is regarded as a passive two-terminal analog device, even for battery-aware computing
– Most batteries being used nowadays are multi-cell battery
• Reality is – Battery is dynamic and complex – Battery operating time and life time
depends on how it is used– Battery may be the least understood
element of a battery-powered computing system, especially for multi-cell battery
Prof. Song Ci, University of Nebraska-Lincoln, [email protected]. All Rights Reserved.
Our Approach: Bring Intelligence into Multi-Cell Battery
• Basic idea: dynamically reconfigure the cell topology of a multi-cell battery based on the load requirements imposed by system and application
Prof. Song Ci, University of Nebraska-Lincoln, [email protected]. All Rights Reserved.
Key Issue: Modeling Adaptive Multi-Cell Battery
Modeling
Temporal-Spatial Behavior Study
1 2 1 20( , ) = ( , ,..., , , ,..., )
T
n mJ y T F x x x y y y dt/ = ( , ), (0) = , [1, ]i i i idx dt G x y x c i n
( , ) 0, [1, ]kR x y k l * ( , ) = arg ( , )max
yy c T J y T
11u
21u
1Ku
12u
22u
2Ku
13u
23u
3Ku
11Nu
21Nu
1KNu
1Nu
2Nu
KNu
1( )Kh u
1 11 2( , )h u u
1 21 2( , )h u u
11 2( , )Kh u u
11 2( , )Kh u u
21 2( , )Kh u u
1 2( , )K Kh u u
11( )h u
21( )h u
2 11 2( , )h u u
2 21 2( , )h u u
21 2( , )Kh u u
1 12 3( , )h u u
1 22 3( , )h u u
12 3( , )Kh u u
2 12 3( , )h u u
2 22 3( , )h u u
22 3( , )Kh u u
12 3( , )Kh u u
22 3( , )Kh u u
2 3( , )K Kh u u
1 11( , )N Nh u u
1 21( , )N Nh u u
11( , )K
N Nh u u
2 11( , )N Nh u u
21( , )K
N Nh u u
2 21( , )N Nh u u
11( , )K
N Nh u u
21( , )K
N Nh u u
1( , )K KN Nh u u
0
0
0
Scientific Challenges: A Battery’s Perspective
• Understand the temporal-spatial behavior of large-scale multi-cell battery under dynamic power management
– Large cell array: high power applications– Dynamic source and load: renewable energy, hybrid vehicles
• Bridge the gap between multi-cell battery dynamics and system dynamics in power management
– Battery and HW such as DC-DC– Battery and SW such as battery-aware task scheduling
• Mission-critical environment, e.g. data center and battlefield• Resource-limited scenarios, e.g. battery-powered embedded systems
• Build a theoretical foundation for application-centric battery-driven power management
– Systematic modeling and holistic optimization– Quantify various design tradeoffs in battery-driven power management
• Among computation , communication, task scheduling, service quality, and battery– Service quality should be the ultimate design goal for power management
Prof. Song Ci, University of Nebraska-Lincoln, [email protected]. All Rights Reserved.