1. Introduction | Model | Solution | Evaluation
Scheduling with Task Duplication
for Computation Offloading
Arani Bhattacharya
Ansuman Banerjee
Pradipta De
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation

2. Computation Offloading
Multiple Applications
Slower Processor
More complex applications in less time
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

3. How Offloading Works Network IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

4. What is the Problem? Scheduler decides how to partition
Partitioning graph is NP-Hard
Applications take time to partition
How to split applications in less time?
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

5. Our Contribution Fast Algorithm Optimal Schedule
Evaluated with synthetic and real workload
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

6. Formal Model Network: Fixed latency Architecture:
One Mobile and one server
Multiple cores
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

7. Formal Model DAG Only local execution allowed
Local or cloud execution allowed
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

8. Minimize application finish time
OBJECTIVE
Minimize application finish time
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

9. Conventional Approach
Uses graph partitioning
High time complexity
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

10. Our Approach Duplication Dynamic Programming algorithm possible
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

11. Example M C 40 X v3 M C 10 X M C 30 24 M C 70 54 M C 74 X v1 v5 v6 v2 M C 50 28
Mobile
Cloud v1 10 v2 20 4 v3 v4 v5 v6 v4
Dependency Migration Time : 10
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

12. Evaluation Simulation-based Trace-based Random graphs
Experiments repeated 1000 times for high confidence interval
Trace-based
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

13. Evaluation Simulation-based Trace-based Random graphs
Experiments repeated 1000 times for high confidence interval
Trace-based
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

14. Makespan IEEE CCNC 2017 Introduction | Model | Solution | Evaluation

15. Scheduling Time IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

16. Evaluation Simulation-based Trace-based SPECjvm08 programs
Obtained by aspect-oriented programming
IEEE CCNC 2017
Introduction | Model | Solution | Evaluation
Introduction | Model | Solution | Evaluation

17. Makespan IEEE CCNC 2017 Introduction | Model | Solution | Evaluation

18. Introduction | Model | Solution | Evaluation
Conclusion
Scheduling with task duplication in offloading
Reduces both makespan and scheduling time
Shown using theory and experiments
IEEE CCNC 2017
Conclusion
Introduction | Model | Solution | Evaluation