SEEDS: A SOFTWARE
ENGINEER’S ENERGY
OPTIMIZATION DECISION
SUPPORT FRAMEWORK
Irene Manotas, Lori Pollock, and James Clause
!
University of Delaware
SOFTWARE ENERGY USAGE: WHY CARE?
Reducing energy consumption can:
Extend battery life (mobile devices)
Reduce electricity bills and related costs
(data centers)
Support sustainability
STATE OF THE ART
Software
Engineer
Can the software engineer help?
Source Code
Compiler
Operating System
Hardware
cpu, disk, etc.
Energy Reducing Transformations
Manage Energy Usage
Dynamic Voltage Frequency Scaling
YES, THEY CAN
Many decisions made by developers impact energy usage:
• Algorithms: Increase or decrease energy usage up to 30% (Bunse et al., 2009; Zhuo and Chakrabarti, 2008)
• Design Patterns:
Increase or decrease energy usage up to 700% (Sahin et al., 2012)
• Refactorings:
Increase or decrease energy usage up to 7% (Sahin et al., 2014)
HOW CAN DEVELOPERS HELP?
Collection x = new ??! Choose the most energy
List x = new ??
efficient implementation
LinkedHashSet
TreeSet
HashSet
EnumSet
CopyOnWriteArraySet
UnifiedSet
UnifiedMap
SetAdapter
MultiReaderFastList
InmutableUnifiedMap
~ 10
CharArraySet
ObjectArraySet
ObjectAVLTreeSet
OpenObjHashBigSet
ObjectRBTreeSet
...
...
~ 40
...
...
~ 40
FastSet
FastBitSet
LocalMap
FastMap
FastList
~ 700
JCF
Hundreds of possibilities for each choice.
Collection x = new ??
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
~ 700
...
...
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
~ 40
~ 40
~ 10
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
~ 700
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
~ 40
~ 40
Hundreds of decision points!
Choose the most energy
efficient implementation
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
~ 10
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
...
~ 700
Collection x = new ??
...
~ 10
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
Hundreds of possibilities for each choice.
...
~ 40
Hundreds of possibilities for each choice.
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
HOW CAN DEVELOPERS HELP?
...
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
...
...
~ 700
Choose the most energy
efficient implementation
...
Collection x = new ??
...
~ 10
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
Hundreds of possibilities for each choice.
HOW CAN DEVELOPERS HELP?
...
...
~ 40
...
...
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
JCF
...
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
Choose the most energy
efficient implementation
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
...
Collection x = new ??
~ 700
efficient implementation
...
Hundreds of possibilities for each choice.
~ 10
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
...
...
JCF
HOW CAN DEVELOPERS HELP?
JCF
~ 10
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
Hundreds of possibilities
for each
choice.
Choose
the most
energy
...
~ 40
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
~ 40
...
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
~ 40
Collection x = new ??
efficient implementation
...
~ 700
JCF
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
...
~ 10
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
...
~ 40
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
HOW CAN DEVELOPERS HELP?
Hundreds of possibilities
for each
choice.
Choose
the most
energy
...
...
...
...
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
JCF
HOW CAN DEVELOPERS HELP?
Collection x = new ??
~ 700
efficient implementation
...
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
~ 700
JCF
...
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
efficient implementation
~ 10
...
...
Hundreds of possibilities
for each
choice.
Choose
the most
energy
Collection x = new ??
~ 40
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
~ 10
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
Hundreds of possibilities
for each
choice.
Choose
the most
energy
...
JCF
HOW CAN DEVELOPERS HELP?
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
~ 40
Collection x = new ??
efficient implementation
...
~ 700
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
...
~ 10
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
HOW CAN DEVELOPERS HELP?
Hundreds of possibilities
for each
choice.
Choose
the most
energy
Collection x = new ??
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
JCF
HOW CAN DEVELOPERS HELP?
...
~ 40
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
...
...
...
...
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
JCF
...
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
efficient implementation
~ 700
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
...
Hundreds of possibilities
for each
choice.
Choose
the most
energy
...
Choose the most energy
efficient implementation
...
HOW CAN DEVELOPERS HELP?
~ 10
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
...
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
Collection x = new ??
...
~ 40
Choose the most energy
efficient implementation
...
~ 700
JCF
Collection x = new ??
UnifiedSet!
UnifiedMap!
SetAdapter!
MultiReaderFastList!
InmutableUnifiedMap
...
~ 10
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
...
~ 40
CharArraySet!
ObjectArraySet!
ObjectAVLTreeSet!
OpenObjHashBigSet!
ObjectRBTreeSet
...
...
...
...
~ 40
FastSet!
FastBitSet!
LocalMap!
FastMap!
FastList!
Collection x = new ??
HOW CAN DEVELOPERS HELP?
...
LinkedHashSet!
TreeSet!
HashSet!
EnumSet!
CopyOnWriteArraySet
Choose the most energy
efficient implementation
HOW CAN DEVELOPERS HELP?
...
HOW CAN DEVELOPERS HELP?
~ 700
JCF
JCF
Hundreds of possibilities for each choice.
Hundreds of possibilities for each choice.
SEEDS
Software Engineer’s Energy-optimization Design
Framework
!
Help developers improve energy usage of their
applications by making decisions about which
source-level changes to apply
THE SEEDS FRAMEWORK
Application
SEEDS
Energy Optimized
Modifications
1. Automatically apply changes to explore decision
space with regard to energy usage (many versions)
2. Abstract away tedious system level concerns:
create versions; monitor energy usage; select 3. Support different software engineering decisions
SEEDS APPROACH
Inputs
Application Code
Potential Changes
Optimization
Parameters
Context
Information
Outputs
Energy Optimized
Application
Modifications
!
Define
ApplicationSpecific
Search Space
Search
Optimization
Results
SEEDS FRAMEWORK COMPONENTS
Inputs
Outputs
Search
Application Code
Potential Changes
Optimization
Parameters
Context
Information
!
Define
ApplicationSpecific
Search Space
Select Solution
Transform
Application
Profile Energy
Usage
Energy Optimized
Application
Modifications
Optimization
Results
AN INSTANTIATION: SEEDSAPI
SEEDSapi supports the selection of Collections
implementations to optimize Java applications.
Java Application
Collections
Libraries
Implementations
Set, List
Application Test
Suite
Define
ApplicationSpecific Search
Space
Application
Generalization
Identify
Collection
Allocations
Filter choices
based on
optimization
parameters.
Search
Select Implementation
Transform
Application
Profile Energy
Usage
Optimized Java
Application
Optimization
Results
EVALUATION OF SEEDS
• RQ1: Effectiveness
Can SEEDS improve the energy usage of applications?
• RQ2: Exploration Capability
How does the size of the search space impact effectiveness?
How often do developers choose the most energy efficient
implementation?
How often is an implementation the most energy efficient?
• RQ3: Cost
What is the time required to find the best solution?
EVALUATION METHODOLOGY
• Independent Variable: Library Implementation
157 implementations from 6 Collections Libraries
• Dependent Variable: Energy Usage (joule)
• Subjects
Seven Java applications
8k - 100k LoC
JCF
LEAP
Machine
Barbecue, Jdepend,
Apache-XML-Security,
Jodatime, Commons-Lang,
Commons-beanutils,
Commons-cli
RQ1: EFFECTIVENESS
Application
Barbecue
Jdepend
Apache-xml-security
JodaTime
Commons-lang
Commons-beanutils
Commons-cli
% Improvement
(JCF only)
17*
3*
5
8*
10
—
2*
% Improvement
(all 6 libraries)
17*
6*
5
9
13
—
2*
Interactions between transformations cancel out
expected benefits
* only one change
†
multiple changes
SEEDSapi was able to improve the energy
efficiency of 6/7 applications.
RQ2: EXPLORATION CAPABILITY
How does the size of the search space
impact effectiveness?
Application
Barbecue
Jdepend
Apache-xml-security
JodaTime
Commons-lang
Commons-beanutils
Commons-cli
* only one change
% Improvement
(JCF only)
17*
3*
5
8*
10
—
2*
†
% Improvement
(all 6 libraries) Increase
17*
6*
3
5
9
1
13
3
—
2*
-
multiple changes
More choices can result in larger improvements.
RQ2: EXPLORATION CAPABILITY
How often do developers choose the most
energy efficient implementation?
• SEEDS analyzed a total of 123 decision points
• Switching from the developer’s original choice to a
new choice improved the energy usage:
JCF only: 56% (69 times)
All Libraries: 72% (89 times)
Developers are unlikely to choose the most
efficient implementation.
RQ2: EXPLORATION CAPABILITY
How often is a given implementation the most
energy efficient?
Selection Frequency
12%
8%
4%
0%
No single implementation is the most Collections API Implementation
energy
efficient.
RQ3: COST OF SEEDS
Optimization time (hours)
250!
Other Parts of SEEDS!
Profiling Energy!
200!
150!
100!
50!
0!
JCF! ALL!
JCF! ALL!
Barbecue!
Jdepend!
Cost is reasonable
JCF! ALL!
JCF! ALL!
JCF! ALL!
JCF! ALL!
JCF! ALL!
Apachexmlsecurity!
Jodatime!
Commonslang!
Commonsbeanutils!
Commonscli!
Cost could be improved by using different
energy measurement techniques
CONCLUSIONS
Designed SEEDS, a general framework for supporting
decisions to improve applications’ energy usage.
• Demonstrated feasibility/usefulness:
• SEEDSapi, an instantiation of SEEDS for API
implementation selection.
• Evaluated SEEDS with 7 Applications:
- effective at improving the energy usage of Java
applications
- supports asking and answering energy-usage related
questions •
FUTURE WORK
•
•
Improve SEEDS search strategy
•
many decisions
•
many decision points in code
Explore more instantiations •
how general
IMPROVING SEEDS SEARCH
• What search-based techniques make the most
sense?
• What properties does the search space have?
• How do we frame the problem to use a searchbased technique?
ADDITIONAL RELATED WORK
• Measure the Energy Consumption of Applications
Provide energy usage of application by different approaches
(HW instrumentation, simulation, estimation) (e.g., Singh, 2010;
Hao, 2013).
Does not provide information about which
changes could be made to reduce the energy.
• Language Support
Implement energy management policies according to device
status, tasks being executed, events or features (e.g., Cohen,
2012; Malakuti, 2014; Siegmund, 2010).
Need to know which policies to implement and
where and when to apply them manually.