1. BENJAMIN BILLET and VALERIE ISSARNY
SPINEL: An Opportunistic Proxy for Connecting Sensors to the Internet of Things
ACM Transactions on Internet Technology, Vol. 17, No. 2, March 2017
BENJAMIN BILLET and VALERIE ISSARNY
September 2017
March 2017

2. OUTLINE Background The Spinel Opportunistic Proxy Conclusion
Assessment
Conclusion

3. Background Background
The new generation of smart Things (6LowPAN, CoAP, RESTful)
The existing WSNs rely on resource-constrained motes and various proprietary communication networks and protocols.
Deploying these new standards is a difficult task for existing sensor networks, as each device must be updated or replaced.
The network layer are not homogeneous.
Setting up a gateway
The complexity and the cost
Each sensor’s owner has to buy, install, and manage these devices

4. Background
The paper promoted an alternative approach to the networking of Things by leveraging smartphones as opportunistic proxies that discover nearby static sensors while moving, register them to the IoT discovery system, and forward the users’ queries to the related sensors.

5. Background
Problems
Due to the short connectivity ranges of involved wireless technologies, the set of nearby sensors can vary quickly while the smartphone is moving.
The smartphone spends a lot of energy on discovering and registering unnecessary sensors, inducing a strong overload by sending multiple updates to the discovery system, regardless of its level of decentralization.
In addition, the same problem affects the queries sent to the network, as smartphones can drift away from the sensors while a query is processed.

6. Background Requirements
When the smartphone is not moving at a given location (a pause), the opportunistic proxy needs to
(i) infer when it is going to move again and
(ii) check if the user will stay at the same place long enough to bear the cost of discovery and registration.
In order to reduce the overall energy consumption, the opportunistic proxy should limit the number of interactions with the discovery system to minimize the active time of the wireless interfaces.

7. The Spinel Opportunistic Proxy
Interactions with the IoT Infrastructure

8. The Spinel Opportunistic Proxy
Architecture Design
Both internal and external sensors are abstracted as
virtual sensors(HAL?).
This avoids the need to modify the existing discovery systems to support proxies.
The number of devices to register is reduced as only the smartphone is registered instead of the entire set of nearby sensors.
the virtual sensor abstraction can be used to create software sensors based on various continuous data sources.

9. The Spinel Opportunistic Proxy
Architecture Design

10. The Spinel Opportunistic Proxy
Architecture Design
In order to reduce the energy consumption,
SPINEL analyzes when and how the smartphones’ mobility sensors can be used, depending on their respective energy costs and the required accuracy
SPINEL continuously builds a mobility database that stores past locations and paths followed by the user over time

11. The Spinel Opportunistic Proxy
Mobility Analysis

12. The Spinel Opportunistic Proxy
Mobility Analysis
SPINEL analyses the smartphone mobility to infer where and when to discover nearby sensors and register them to the discovery system. The mobility analysis process comprises three steps:
—Check if the smartphone is moving, using the embedded mobility sensors by their increasing order of energy-consumption.
—Acquire the most accurate location of the smartphone (less than 5m, typically provided by the GPS sensor) if not moving.
—Infer how long the smartphone will stay at this location, given past pause times.

13. The Spinel Opportunistic Proxy
Mobility Analysis
SPINEL acquires the precise position of the user :
—If the position is in the mobility database, they use one of the following strategies: (i) the mobility process detects that the user is static x consecutive times, (ii) the nature of the location or the time of the day is statistically related to a particular pause time, or (iii) a statistical predictor can be used.
—If the position is not in the mobility database, they use the past pause times at the same location to trigger the other processes.

14. The Spinel Opportunistic Proxy
Path Prediction
Estimation of Path Similarity

15. The Spinel Opportunistic Proxy
Path Prediction
Estimation of Path Similarity
|P| = |Q|, in which case the similarity of P and Q is computed by aggregating the individual similarities of pauses (pi, qi) ∀i ∈[1, |P|].
|P| ǂ |Q|, in which case we must select which pairs (pi, qj ) are going to be evaluated.

16. The Spinel Opportunistic Proxy
Path Prediction
Estimation of Path Similarity

17. The Spinel Opportunistic Proxy
Path Prediction
Estimation of Path Similarity
SPINEL uses strict aggregation to check whether two paths must be merged in the mobility database and soft aggregation for finding a past path that matches the current path followed by the user.
After the execution of the algorithm, all the k = |P| + |Q| − 2|M|; similarity values S = (s1, , sk) are aggregated for computing the similarity value of P and Q.

18. The Spinel Opportunistic Proxy
Path Prediction
Mobility Database Construction and Lookup

19. The Spinel Opportunistic Proxy
Path Prediction
Mobility Database Construction and Lookup
In order to ensure that the mobility database does not grow too big, SPINEL limits the size of each cluster by applying an eviction policy for the stored paths that are either too old (LRU strategy) or not followed frequently enough by the user (LFU strategy).
If there is one or more candidate paths with a soft similarity score greater than a threshold β, then the path with the highest score and time length is selected and sent to the registry. The purpose of such a threshold is to reduce the number of wrong predictions sent to the registry.

20. The Spinel Opportunistic Proxy
Discussion on Security, Privacy, and Incentives
Security & Privacy
The registry is trusted (similarly to a DNS server) and authenticated, providing strong and reliable encryption for communicating with the smartphones.
Additional anonymization techniques is used.
The proxy provides additional mechanisms (encryption, pre-aggregation, etc.).
Incentives
Directly improving the quality of the information
Altruistic or democratic incentive
Gamification
Reward

21. Assessment
The prototype of SPINEL

22. Assessment The prototype of SPINEL Connectors
A connector for the MobIoT path-aware sensor registry
A connector for the Dioptase data streaming and continuous processing
The proxy is a background daemon, which monitors the mobility using the mobility analysis process
The discovery is performed by using the standard Android connectivity APIs:
Bluetooth Service Discovery Protocol
DNS-SD for Wi-Fi/Wi-Fi Direct
Continuous NFC discovery

23. Assessment Evaluate Two Aspects of the Opportunistic Proxy
The impact of the mobility analysis process on the energy consumption compared to GPS- and accelerometer-only approaches
The amount of messages exchanged between SPINEL proxies and registries, which must be low in order to reduce both the network load and the active time of the smartphones’ wireless interfaces, thereby also limiting the energy consumption

24. Assessment
Energy Consumption of the Mobility Analysis Process

25. Assessment
Network Message Load

26. Conclusion Contributions
—SPINEL reduces the impact of network heterogeneity and integrates existing sensor networks within the IoT without deploying new static gateways or modifying the existing infrastructure.
—SPINEL alleviates the negative impact of mobility using two mechanisms.
The process selects the best location sensors for inferring whether the user is moving, depending on the required accuracy and energy consumption.
The path prediction technique, called Complete Path Prediction (C2P), analyzes and anticipates users’ followed paths.

27. Conclusion Technical Perspective
Investigate the relevance of using a human mobility model to improve results of prediction process
Extend the opportunistic proxy from static sensors to mobile sensors
Investigate how to infer automatically the values of the thresholds and constants used
Integrate indoor positioning techniques in order to apply our prediction algorithm to areas where the GPS is not available

28. Conclusion 个人感想 利用手机作为机会代理，本文对已有的传感器网络系统提出了一种集成解决方案
本文对群智感知的概念有扩展：从个人手机传感器众包到手机作为机会代理众包附近的传感器
有些问题交代不清楚：如有关传感器连接的相关问题交代不明确：如Virtual Sensor，Software Sensor，以及其如何即插即用（PnP，对智能对象的PnP有帮助）？又如有关算法未详细交代，human activity recognition 和transportation detection？
感觉有两个地方有问题：一个是公式；另一个是算法

29. Thank you!
Q&A
That’s all, thank you for your attention.