Adaptive Code Offloading for Mobile
Cloud Applications
Exploiting Fuzzy Sets and Evidence-based Learning
Huber Flores
hube...
Outline
• Background
• Problem statement
• Proposed solution
• Conclusions
MCS'13, Taipei, Taiwan 2
Background
• Mobile cloud computing
– Augmented functionality
– Extended battery life
– Increased performance
• Task deleg...
Background
MCS'13, Taipei, Taiwan 4
Background
• MAUI (.Net)
– Code annotations (Cuervo et al., 2010)
• CloneCloud
– Code profilers (Chun et al., 2011)
• Othe...
What is the problem?
Besides, scalability…
MCS'13, Taipei, Taiwan 6
Problem statement
• Code offloading may also fail?
– Runtime analysis should be encouraged (Ra et al.,
2011)
– Some code c...
Problem statement
• Is Mobile Cloud taking full advantage of Cloud
Computing?
• Code offloading for next generation mobile...
Proposed solution
• Offloading from a different perspective
– “Offloading is a global learning process rather than
local d...
Evidence-based mobile code offloading
MCS'13, Taipei, Taiwan 10
Evidence-based mobile code offloading
MCS'13, Taipei, Taiwan 11
Evidence-based mobile code offloading
• Crisp Sets
– Mobile parameters
– Cloud parameters
– Others
• Linguistic variables
...
Preliminary results
• Use cases
– Mobile components scheduling
– Back-end allocation (Clone)
– Data-intensive REST
– Other...
Preliminary results
Bandwidth = speed_high
Data size = data_small
Instance load = cpu_highOffload = 90%
Not to offload = 6...
Preliminary results
Bandwidth = speed_high
Data size = data_small
Instance load = cpu_normalOffload = 90%
Offload = 83%
MC...
Preliminary results
Bandwidth = speed_high
Data size = data_medium
Instance load = cpu_normal
Not to offload = 70%
Not to ...
Preliminary results
Bandwidth = speed_high
Data size = data_medium
Instance load = cpu_normal
Offload = 75%
Instance type ...
Preliminary results
MCS'13, Taipei, Taiwan 18
Conclusions and future research
• Cloud analysis may periodically empower
mobile components with knowledge. (Cloud is
expe...
THANK YOU FOR LISTENING…
MCS'13, Taipei, Taiwan 20
Upcoming SlideShare
Loading in …5
×

Adaptive Code Offloading for Mobile Cloud Applications

1,237 views

Published on

Check the complete paper: http://dl.acm.org/citation.cfm?id=2482984

Published in: Education, Travel, News & Politics
0 Comments
1 Like
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
1,237
On SlideShare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
0
Comments
0
Likes
1
Embeds 0
No embeds

No notes for slide

Adaptive Code Offloading for Mobile Cloud Applications

  1. 1. Adaptive Code Offloading for Mobile Cloud Applications Exploiting Fuzzy Sets and Evidence-based Learning Huber Flores huber@ut.ee MCS'13, Taipei, Taiwan 1
  2. 2. Outline • Background • Problem statement • Proposed solution • Conclusions MCS'13, Taipei, Taiwan 2
  3. 3. Background • Mobile cloud computing – Augmented functionality – Extended battery life – Increased performance • Task delegation – Mobile Cloud Middleware (Flores et al., 2011) • Zompopo (Srirama et al., 2011) • Code offloading – Small data size is transmitted which requires intensive computational processing (Kumar et al., 2010) – Online/Offline MCS'13, Taipei, Taiwan 3
  4. 4. Background MCS'13, Taipei, Taiwan 4
  5. 5. Background • MAUI (.Net) – Code annotations (Cuervo et al., 2010) • CloneCloud – Code profilers (Chun et al., 2011) • Other frameworks – Own framework (Similar to MAUI, but Java-based) MCS'13, Taipei, Taiwan 5
  6. 6. What is the problem? Besides, scalability… MCS'13, Taipei, Taiwan 6
  7. 7. Problem statement • Code offloading may also fail? – Runtime analysis should be encouraged (Ra et al., 2011) – Some code cannot be profiled (e.g. REST) – Should it be a local decision of global inference • Cloud infrastructure MCS'13, Taipei, Taiwan 7
  8. 8. Problem statement • Is Mobile Cloud taking full advantage of Cloud Computing? • Code offloading for next generation mobile devices? e.g., Samsung Galaxy S, S2, S3, S4…. (How to optimize the offloading decision process?) MCS'13, Taipei, Taiwan 8
  9. 9. Proposed solution • Offloading from a different perspective – “Offloading is a global learning process rather than local decision process“ • How it can learn? – Analysis of code offloading traces which are generated by the massive amount of devices that connect to cloud “EMCO: Evidence-based mobile code offloading“ MCS'13, Taipei, Taiwan 9
  10. 10. Evidence-based mobile code offloading MCS'13, Taipei, Taiwan 10
  11. 11. Evidence-based mobile code offloading MCS'13, Taipei, Taiwan 11
  12. 12. Evidence-based mobile code offloading • Crisp Sets – Mobile parameters – Cloud parameters – Others • Linguistic variables – bandwith • Fuzzy Sets – speed_slow, – speed_normal – speed_high • Variable to control – Offloading • Rules – if speed_high AND data_small then remote – If speed_low AND data_medium the local MCS'13, Taipei, Taiwan 12
  13. 13. Preliminary results • Use cases – Mobile components scheduling – Back-end allocation (Clone) – Data-intensive REST – Others MCS'13, Taipei, Taiwan 13
  14. 14. Preliminary results Bandwidth = speed_high Data size = data_small Instance load = cpu_highOffload = 90% Not to offload = 65% MCS'13, Taipei, Taiwan 14
  15. 15. Preliminary results Bandwidth = speed_high Data size = data_small Instance load = cpu_normalOffload = 90% Offload = 83% MCS'13, Taipei, Taiwan 15
  16. 16. Preliminary results Bandwidth = speed_high Data size = data_medium Instance load = cpu_normal Not to offload = 70% Not to offload = 75% Instance type = micro Instance cores = 1 Battery level = low MCS'13, Taipei, Taiwan 16
  17. 17. Preliminary results Bandwidth = speed_high Data size = data_medium Instance load = cpu_normal Offload = 75% Instance type = medium Instance cores = 2Not to offload = 82% Battery level = low MCS'13, Taipei, Taiwan 17
  18. 18. Preliminary results MCS'13, Taipei, Taiwan 18
  19. 19. Conclusions and future research • Cloud analysis may periodically empower mobile components with knowledge. (Cloud is expert and handset asks for its expertise) – e.g. code offloading • Offloading as a learning process may grant the cloud with self-adaptive behavior. • We are exploring some other strategies for the analysis of code offloading traces. MCS'13, Taipei, Taiwan 19
  20. 20. THANK YOU FOR LISTENING… MCS'13, Taipei, Taiwan 20

×