Download - ENDA - Presentation - MCC workshop - v1.11
![Page 1: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/1.jpg)
ENDA: Embracing Network Inconsistency for Dynamic Application Offloading
in Mobile Cloud ComputingJiwei Li Kai Bu Xuan Liu Bin Xiao
The Hong Kong Polytechnic University
Presenter: Jiwei Li
![Page 2: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/2.jpg)
Outline
• Background & Problem• Proposed Solution• Preliminary Results• Conclusion
![Page 3: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/3.jpg)
Outline
• Background & Problem• Proposed Solution• Preliminary Results• Conclusion
![Page 4: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/4.jpg)
Mobile Cloud Computing
• Applications– Apple’s iCloud, Dropbox
• Technical problems – MCC architecture & infrastructure– Network connectivity– Energy efficiency
• One important research topic - offloading
![Page 5: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/5.jpg)
Offloading Strategy
• Previous work– MAUI, CloneCloud, Odessa, COMET
![Page 6: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/6.jpg)
Offload to Cloud
Compute-intensive applications
Computed results
High latency (100-300ms)Limited bandwidth (386 Kbs to 3.6 Mbs)High energy consumption
Nearly unlimited resources
![Page 7: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/7.jpg)
Offload to Cloudlet
Compute-intensive applications
Computed results
Low latency (23-50ms)High bandwidth (54 Mbs)
Limited coverage of Wi-Fi (20-100m)Resource constraint
![Page 8: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/8.jpg)
Uninvestigated Issues in Offloading
• Offloading at mobile environments
• Balancing workloads among multiple cloudlets
Our research is focused onoffloading to cloudlets through Wi-Fiat mobile environments.
![Page 9: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/9.jpg)
A B
C
D
A Motivating Example
![Page 10: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/10.jpg)
Re-connection Matters
• Re-connection includes– Scanning– Connecting– Assigning IP and network ID
• Takes long time (1-12s)• Consumes additional power
Reducing re-connection times means increasing energy efficiency.
![Page 11: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/11.jpg)
Our Studied Problems
• How to predict user’s trajectory?• How to select Wi-Fi access points (AP)?• How to balance workload among cloudlets?
![Page 12: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/12.jpg)
Problem Formulation
• Minimize: – Communication overheads during offloading at
mobile environments• Must satisfy requirements:– App-specific network latency– App-specific response time
To put it simply, we aim toselect the most energy-efficient Wi-Fi access point,taking user mobility and server load into account.
![Page 13: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/13.jpg)
Outline
• Background & Problem• Proposed Solution• Preliminary Results• Conclusion
![Page 14: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/14.jpg)
Answering a few questions …
• Is it feasible to deploy cloudlets at large scale?• Bind current public Wi-Fi hotspots with cloudlets.
•How do we overcome resource constraints on cloudlets?• Adopt workload balance management mechanism among
participating cloudlets.
•How do we conquer Wi-Fi’s limited coverage range issue?• Propose mobility-aware Wi-Fi AP selection scheme.
![Page 15: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/15.jpg)
A Real Scenario
![Page 16: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/16.jpg)
ENDA
• Three-tier architecture Design– Cloud– Cloudlet– Smartphone
• Objective:– Make the most energy efficient offloading decision
![Page 17: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/17.jpg)
Clouds
CloudletsSmartphones
VM on cloudlets
Profilers
Wi-Fi
2G/3
G
WAN
User Track Prediction
Wi-Fi AP Distribution and Status
Wi-Fi AP Selector
GPS
Runtime System
Wi-Fi Adapter
FINAL DECISION
OFFLOADING
INPUT INPUTRE
PORT
REPO
RT
APP
INFO
Our work will be focused on
![Page 18: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/18.jpg)
Advantages
• Minimize end-to-end communication overheads
• Exempt smartphones from complex computation of making decisions
• Improve energy efficiency for offloading
![Page 19: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/19.jpg)
Demo Scenario
Predicted user track(will be pruned based onapp info & network conditions)
Effective routes:N1 -> (S, A)N2 -> (S, B)N3 -> (S, D)N4 -> (C, D)
ENDA chooses the most energy-efficient Wi-Fi AP according to the specific predicted track
Start offloading at location S
![Page 20: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/20.jpg)
Outline
• Background & Problem• Proposed Solution• Preliminary Results• Conclusion
![Page 21: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/21.jpg)
GUI-based Simulation
Add routers
Add walking path
Calculate effective path
![Page 22: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/22.jpg)
Simulation Results
Wi-Fi B Wi-Fi A Wi-Fi C
![Page 23: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/23.jpg)
Outline
• Background & Problem• Proposed Solution• Preliminary Results• Conclusion
![Page 24: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/24.jpg)
Conclusion
• ENDA– Difference from previous work– Minimize communication overheads– Potential to apply to real offloading systems
• Future work– Thorough mathematical analysis– Implementation– More complex scenarios
![Page 25: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/25.jpg)
Thank you!
![Page 26: ENDA - Presentation - MCC workshop - v1.11](https://reader035.vdocuments.mx/reader035/viewer/2022081503/587ebc4c1a28abbb688b6ed1/html5/thumbnails/26.jpg)
Q&A