1. Energy-Aware Opportunistic Mobile Data Offloading
Sylvia Kouyoumdjieva and Gunnar Karlsson
ITC28, COST Action ACROSS on "Quality Engineering for a Reliable Internet of Services" 
Würzburg,

2. Opportunistic Mobile Data Offloading
Goal: Minimize traffic volumes in the cellular network.
Assumption 1: Nodes have infinite battery capacity!
Assumption 2: Nodes are altruistic!

3. Opportunistic Mobile Data Offloading (2)
Goal: Deliver satisfactory application throughput at a low energy price.
Realistic assumption 1: Nodes have limited battery capacity!
Realistic assumption 2: Nodes are selfish!

4. Data-seeking vs Data-fulfilled Nodes
Data-seeking node:
Node that is still in search for contents
Objective: Obtain all content items
at a low energy cost
Data-fulfilled node:
Node that has obtained all content items
Objective: Contribute to data dissemination
at a low(est) energy cost

5. Data-seeking (Searches for data)
Data-seeking Nodes
Duty cycling [1]
Cycling interval Tc
Listening duration d ~ Uni(0, Tc)
Data-seeking (Searches for data)
[1] S.T. Kouyoumdjieva, G. Karlsson,  Impact of Duty Cycling on Opportunistic Communication, in Mobile Computing, IEEE Transactions on, vol.15, no.7, pp , July 2016

6. Data-fulfilled Nodes Strict Selfishness
Node opts out of dissemination process after it obtains all content items.
Progressive Selfishness [2]
Node behaves in a progressively selfish manner after it obtains all content items.
w, inactivity window
[2] S.T. Kouyoumdjieva, G. Karlsson, Energy-aware opportunistic mobile data offloading under full and limited cooperation, in Elsevier Computer Communications, vol. 84, pp , June 2016

7. Is Duty Cycling Really Needed?
Configuration
Energy
Goodput
ON-SS, p = 10%
56.3%
86.2%
DC-SS, p = 10%
30.4%
82.8%
ON-SS, p = 100%
20.8%
98.3%
DC-SS, p = 100%
11.5%
97.4%
+26%
-4%
ON-SS Nodes are always listening while data-seeking then become strictly selfish when data-fulfilled
DC-SS Nodes duty cycle while data seeking then ditto
p – percentage of participating nodes in dissemination process
[3] S.T. Kouyoumdjieva, G. Karlsson, The Virtue of Selfishness: Device Perspective on Mobile Data Offloading, in Proc. IEEE WCNC 2015

8. Evaluation Scenarios: Mobility
Subway Station
Östermalm
Characteristics:
12 entry points
High mobility scenario
Area: 5872 m2
Characteristics:
5 entry points
Burstiness and queuing
Area: 1921 m2
Legion Studio – designing and dimensioning large-scale spaces via simulation of pedestrian behaviors
Traces are available at:

9. Evaluation Scenarios: Metrics
Goodput (Application throughput)
How much data is offloaded?
Energy Consumption
How much energy is consumed in comparison to a direct cellular download?
Initial content availability q
Amount of data initially injected into a node’s cache
Injection probability pi
Probability that a node obtains contents by the operator upon entering the area

10. Evaluation Scenarios: Configurations
ON – Nodes do not apply any energy saving mechanisms
DC – Nodes duty cycle while data-seeking and data-fulfilled
DC-SS – Nodes duty cycle while data-seeking and become strictly selfish when data-fulfilled
DC-PS – Nodes duty cycle while data-seeking and become progressively selfish when data-fulfilled

11. Evaluation Scenarios: Cooperation
Full Cooperation
All nodes in the system follow the progressive selfishness algorithm.
Limited Cooperation
A subset of nodes are entirely egocentric; non- cooperative nodes only consume resources of other peers in the network without contributing to the data dissemination.

12. Performance under Full Cooperation

13. Östermalm: Effect of Node Density (1)
* S.T. Kouyoumdjieva, G. Karlsson, Energy-Aware Opportunistic Mobile Data Offloading for Users in Urban Environments, in Proc. IFIP/TC6 Networking, Toulouse, France, May 2015

14. Enhanced Progressive Selfishness (1)

15. Enhanced Progressive Selfishness (2)

16. Östermalm w/ λ = 0.01 n/s: Effect of Initial Content Availability (1)

17. Östermalm w/ λ = 0.15 n/s: Effect of Initial Content Availability (2)

18. Subway: Effect of Injection Probability

19. Performance under Limited Cooperation

20. Östermalm w/ λ = 0.01 n/s: Effect of Node Density (1)

21. Östermalm w/ λ = 0.15 n/s: Effect of Node Density (2)

22. Östermalm w/ λ = 0.15 n/s: Effect of Injection Probability (1)

23. Östermalm w/ λ = 0.15 n/s: Effect of Injection Probability (2)
Goodput
Energy consumption

24. Conclusions
We examined the performance in terms of goodput and energy consumption of direct D2D communication in the context of opportunistic mobile data offloading with nodes under full and limited cooperation.
We introduced progressive selfishness on top of duty cycling.
Energy savings from opportunistic communication up to 90% without significant loss in goodput.
Total energy consumption comparable to that consumed by nodes when downloading data _directly_ from the cellular network.
Tolerance for up to 50% of non-cooperative nodes without significantly deteriorating goodput and energy consumption at a system level.

25. References
[1] S.T. Kouyoumdjieva, G. Karlsson,  Impact of Duty Cycling on Opportunistic Communication, in Mobile Computing, IEEE Transactions on, vol.15, no.7, pp , July 2016
[2] S.T. Kouyoumdjieva, G. Karlsson, Energy-Aware Opportunistic Mobile Data Offloading under Full and Limited Cooperation, in Elsevier Computer Communications, vol. 84, pp , June 2016
[3] Ó. Helgason, S.T. Kouyoumdjieva, G. Karlsson, Opportunistic Communication and Human Mobility, in Mobile Computing, IEEE Transactions on, vol.13, no.7, pp , July 2014
[4] Mobility traces: