Application Transformations for Energy and Performance-Aware Device Management - PowerPoint PPT Presentation

About This Presentation
Title:

Application Transformations for Energy and Performance-Aware Device Management

Description:

Application Transformations for Energy and Performance-Aware Device Management Taliver Heath, Eduardo Pinheiro, Jerry Hom, Ulrich Kremer, and Ricardo Bianchini – PowerPoint PPT presentation

Number of Views:203
Avg rating:3.0/5.0
Slides: 43
Provided by: Preferr999
Category:

less

Transcript and Presenter's Notes

Title: Application Transformations for Energy and Performance-Aware Device Management


1
Application Transformations for Energy and
Performance-Aware Device Management
  • Taliver Heath, Eduardo Pinheiro, Jerry Hom,
    Ulrich Kremer, and Ricardo Bianchini
  • Rutgers University

2
The Problem
  • Conserve energy in devices
  • Must take advantage of lower power states
  • State transitions have overhead
  • Cost in both energy and performance
  • Challenge non-interactive applications and fast
    processors
  • Short device idle times
  • Devices cannot use lower power states

3
Our Solution
An Unmodified Application (UM)
CPU
Disk
CPU
idle
idle
active
active
Disk
idle
idle
active
active
Transformed Application
4
Our Goals
  • Conserve energy by exploiting transformations
    that increase idle time
  • Evaluate ideas using
  • Hand-modified programs
  • Automated compiler transformations
  • Specific policies Energy Oblivious, Fixed
    Threshold, Direct Deactivation, Pre-Activation,
    and Combined

5
Application Transformations
  • Increase idle times with help of compiler or
    programmer
  • Identify loops where accesses occur
  • Perform loop transformations
  • Estimate device idle times
  • Insert system calls
  • Idle time limited by memory or real-time
    constraints

6
Example Original Application
i 1 while i lt N read chunki of
file compute on chunki i i1
7
Example Transformed Application
available how_much_memory() numchunks
available/sizeof(chunks) compute_time
appfunc(numchunks) i 1 while i lt N read
chunkiinumchunks of file next_R(compute_time
) compute on chunkiinumchunks i
inumchunks
8
Compiler Framework
  • Annotations to file descriptors
  • Replace disk calls using interprocedural analysis
  • Profiling
  • Buffered I/O
  • Notify OS of idle times
  • Based on SUIF infrastructure

9
Device Management
  • Energy-Oblivious (EO)
  • Fixed-Threshold (FT)
  • Direct-Deactivation (DD)
  • Pre-Activation (PA)
  • Combined(CO) DDPA
  • Final state based on model Heath02

10
Sample Disk Power Graphs (mp3 player)
UM
FT
CO
11
Experimental Setup
  • Fujitsu Disk
  • 6-GB, 4200-rpm laptop disk
  • 3 states
  • Idle 0.9 W
  • Standby 0.22 W
  • Sleep - 0.09W
  • Buffer memory available 19MB
  • Time allowed for reading .3 seconds

12
Experiment
  • 6 applications
  • Mp3 player, mpeg-player
  • Gzip, sftp, mpeg-encode, image smoother
  • Variables investigated
  • Disk management policies
  • Compiler vs. hand-optimized
  • OS prefetching on/off

13
Non-streaming SFTP
14
Streaming MP3 player
15
Average Hand-Modified Results
Policy Energy Performance
EO 40 0
FT 60 5
DD 73 7
PA 60 1
CO 70 4
16
Average Compiler Results
Policy Energy Performance
EO 46 1
FT 68 4
DD 79 7
CO 75 3
17
Related Work (partial list)
  • Application-controlled power states
  • Concept, but no implementation Ellis99,Lu99
  • Compiler infrastructure Delaluz01
  • Direct deactivation and preactivation
    Hom01,Heath02
  • Conserving disk energy Douglis94
  • Modifying disk access API Weissel02

18
Conclusions
  • Application transformations
  • 55-89 savings in energy
  • Minimal effect on performance
  • Idle time predictions are difficult
  • Prefetching has little impact
  • Compiler transformations work well
  • As good as hand modifications
  • Generic framework other disks and devices

19
For more information
  • www.darklab.rutgers.edu

20
Technique
  • Create model of disk energy
  • Transform applications
  • Realize model on real disk
  • Predict disk energy usage
  • Measure disk on 4 applications

21
Future Work
  • More disks
  • Other devices
  • Multiple active processes
  • Asynchronous I/O

22
Summary
23
Historical Use of States
  • Change to Lower State during Period of Idleness
  • Fixed-threshold
  • Adaptive/Heuristic
  • OS Hints
  • Based on general knowledge of system

24
Runlength vs. Energy
25
Projected Application Gain
26
Projected Application Gain
27
Overhead for DD
28
Combined (CO)
CPU
idle
idle
active
active
Disk
idle
active
active
idle
29
Parameter Description
Parameter Explanation
Energy consumed by policy pol
CPU time consumed by policy pol
Run-length
Average power consumed in s
Inactivity threshold for s
Average reactivation energy
Average deactivation energy
Average reactivation time
30
Reality Departs from Model
  • Hidden states in several transitions
  • Transition from active to idle
  • Behavior on activation

For CO
31
Experiments
Modified App Runlengths
Application s1 s2 s3
MP3 player 0 0 1
MPEG player 0 0 1
Image smoother 0 0 1
Gzip 0 .36 .64
Sftp 0 0 1
MPEG encoder 0 0.5 0.5
32
Energy, mpg123
33
Energy, sftp
34
Performance, mpg123
35
Performace, sftp
36
Experimental Results
MP3 player
37
Summary
  • direct-deactivation and preactivation (CO)
  • Can save up to 89 of disk energy
  • No performance penalty, except for MPEG player
    (lt10)
  • Just increasing runlengths, we can save up to 50
    energy
  • Error in model can be significant up to 50 for
    the entire application

38
Energy Oblivious(EO)
CPU
idle
idle
active
active
Disk
idle
39
Direct Deactivation(DD)
CPU
idle
idle
active
active
Disk
40
Pre-Activation(PA)
CPU
idle
idle
active
active
Disk
idle
41
Fixed-Threshold(FT)
CPU
idle
idle
active
active
Disk
42
Terminology
Runlength (R)
Time between device accesses by the processor
R
R
Device Time
CPU Time
Blocking device accesses (reads) Single
ready-to-run application
Write a Comment
User Comments (0)
About PowerShow.com