1. TEMPLE
UNIVERSITY
Deadline-Sensitive Mobile Data Offloading via Opportunistic Communications
Guoju Gaoa, Mingjun Xiao∗a, Jie Wub, Kai Hana, Liusheng Huanga
a University of Science and Technology of China, China
b Temple University, USA

2. Outline
Motivation
Problem
Solution
Example
Simulation
Conclusion

3. Motivation Motivation
(1) Mobile users will be 5.5 Billion (70% of global population) by 2020, up from 4.8 Billion in 2015, a compound annual growth rate (CAGR) of 2.6%.

4. Motivation Motivation
(2) Mobile data traffic will reach Exabytes per month by 2020 (the equivalent of 7,641 million DVDs each month), up from 3.7 Exabytes per month in 2015.
Global mobile data traffic will grow 8-fold from 2015 to 2020, a CAGR of 53%.

5. Motivation Motivation
(3) 51% of global mobile data traffic was offloaded in 2015; 55% of global mobile data traffic will be offloaded by

6. Motivation Motivation Existing offloading models:
Offloading based on DTNs
formulated as a target-set selection problem,
incentive mechanism to attract participants,
etc.
Offloading based on WiFi
economic benefits and load balance problem,
maximize profits from the perspective of system,
A mobile user uploads data items onto the cloud side through WiFi networks when it encounters WiFi APs during Time-To-Lives (TTLs) of data items, or via cellular networks when the TTLs of data items expire, respectively.

7. Motivation Motivation Mobile users: WiFi APs uncertain mobility
self-centered (minimizing transmission cost)
data with deadlines
WiFi APs
lower transmission cost than cellular networks
limited capacity due to restricted connect time
A mobile user uploads data items onto the cloud side through WiFi networks when it encounters WiFi APs during Time-To-Lives (TTLs) of data items, or via cellular networks when the TTLs of data items expire, respectively.

8. Problem Problem Data offloading scenario:
A mobile user uploads data items onto the cloud side through WiFi networks when it encounters WiFi APs during Time-To-Lives (TTLs) of data items, or via cellular networks when the TTLs of data items expire, respectively.
A mobile user uploads data items onto the cloud side through WiFi networks when it encounters WiFi APs during Time-To-Lives (TTLs) of data items, or via cellular networks when the TTLs of data items expire, respectively.

9. Problem Problem Offloading model: Problem:
data set : D = {d1, ・ ・ ・ , di, ・ ・ ・ , dn}, di =< si , ti>
WiFi set: W = {w1, w2, ・ ・ ・, wm}, wj =<τj, pj , qj >
Problem:
Given a fixed number of data items and WiFi APs
Minimize the expected data transmission cost
A mobile user uploads data items onto the cloud side through WiFi networks when it encounters WiFi APs during Time-To-Lives (TTLs) of data items, or via cellular networks when the TTLs of data items expire, respectively.

10. Problem
Problem
By taking into consideration the transmission capacity of WiFi APs and deadlines of data items, our problem involves a probabilistic combination of multiple 0-1 knapsack constraints.

11. Solution Solution Adopted greedy strategy including two phases:
1. selecting data offloading operation that increases the offloading utility function value most quickly:
2. selecting the offloading operation, whose data item has the largest data size.

12. Solution
Solution
The FDO Algorithm (offline algorithm): the user makes the data offloading decisions before it encounters any WiFi AP.

13. Solution Solution Approximation ratio of 2:
We use optF to denote the optimal offline offloading strategy of optimization problem;
We analyze the approximation ratio of FDO, and have the following theorem:

14. Solution
Solution
The NDO Algorithm (online algorithm): the data offloading decision is made only when the mobile user encounters the WiFi APs.

15. Solution Solution Competitive ratio of 2:
The competitive ratio is defined as the ratio of optN and our online solution Φ∗.
We have the following theorem:

16. Example
Example
D = {d1, d2, d3, d4}: s1 = 8, t1 = 11, s2 = 6, t2 = 13, s3 = 5, t3 = 17, s4 = 10, t4 =18.
W = {w1, w2}: τ1 = 10, p1 =0.6, q1 = 15, τ2 = 15, p2 = 0.9, q2 = 10.

17. Example
Example
the initial state: feasible offloading operations are determined, according to TTLs of data and appearing time of WiFi APs.

18. Example
Example
Select (d4, w2):

19. Example
Example
Select (d1, w1):

20. Example Example Select (d2, w1):
Now, both remaining capacities of w1 and w2 are smaller than the size of any data items.

21. Example
Example
Deadline constraint and capacity constraint are both satisfied.
Final offloading strategy Φ is obtained:
Φ={(d1,w1), (d2,w1), (d4,w2)}

22. Simulation Simulation Settings: Parameter name Default value Range
Number of data items: n
150
50-250
Number of WiFi APs: m 15
5-25
Average size of data: s
200
Average TTL of data: t
100
Average capacity of WiFi APs: q
3000
Average accessing probability: p
0.2

23. Simulation Simulation Algorithms in comparison: RS (Random Selection)
SA (Sequential Allocation)
Both RS and SA satisfy deadline and capacity constraints.
Metrics:
Total transmission cost
Offloading Ratio

24. Simulation Simulation Results:
total cost and offloading ratio vs. the number of data items

25. Simulation Simulation Results:
total cost and offloading ratio vs. the number of WiFi APs

26. Simulation Simulation
Results: total cost vs. average TTLs and sizes of data items, average capacities and accessing probabilities of WiFi APs

27. Simulation Simulation
Results: offloading ratio vs. average TTLs and sizes of data items, average capacities and accessing probabilities of WiFi APs.

28. Conclusion Conclusion
NDO and FDO algorithms outperforms the compared algorithms in both the total cost and offloading ratio. NDO achieves the best performance, and FDO follows.
When the number of data items or the average size of data increases, the total cost increases and the offloading ratio reduces significantly.
When the number of WiFi APs or the average capacity of WiFi APs increases, the total cost decreases and the offloading ratio increases observably.
When the average TTL of data or the average probability of accessing WiFi APs increases, the cost decreases, while offloading ratio increases, respectively.

29. Thank You! Q&A