Enabling Scalable Publish/Subscribe for Logical-Clustering in Crowdsourcing via MediaSense
- 1. Hasibur Rahman Enabling Scalable Publish/Subscribe for Logical-Clustering in Crowdsourcing via MediaSense Science and Information Conference 2014 August 27-29, 2014 | London UK Department of Computer and Systems Sciences (DSV) Stockholm University
- 2. Overview • Introduction • Background • Motivation • Research Problem • Our Solution • Advantages and Analysis • Conclusions and Future Work
- 3. Introduction • Crowdsourcing • Increase of Information • Vast IoT sources • Logical-Clustering • Efficient Management of Context Information (CI) • Filters out similar context from distributed sources • MediaSense • A scalable distributed IoT platform for CI sharing
- 4. Background • Spontaneous human participation i.e. crowdsourcing is pivotal for future pervasive computing (Franco- 2011) • The surge of social networks, mobile devices, Internet or Web-enabled services have enabled unprecedented level of human participation in crowdsourcing which has been branded as “human-in-the- loop-sensing” or citizen sensor networks (Boulos et al. -2011), Sheth -2009) • Ericsson predicts that 50-500 billion mobile devices will be in use by 2020 (Ericsson - 2013) • Ericsson envisions 5G as enabler for Networked Society (Ericsson - 2013) • This necessitates proper management of CI so that resources can be shared from remote places • Logical-clustering has been proposed as opposed to physical clustering to efficiently manage CI (Rahman et al. - 2013)
- 5. Crowdsourcing • People • Pervasive devices • Internet or Web-enabled services • Surrounding things • Context Information
- 6. Motivation • Sharing heterogenous CI obtained from distributed sources • Publish/Subscribe (PubSub) has emerged as an efficient means of sharing ubiquitous CI • By leveraging the PubSub in the crowdsourcing model can unravel the challenge of sharing CI in real-time
- 7. What is the problem? Can the distributed MediaSense platform (Kanter et al., 2009) be used as scalable PubSub model in real-time? If it does then how does this approach differ from other approaches?" We have proposed MediaSense as a potential solution to the above research questions How can the context-IDs in logical-clustering be shared efficiently in real-time? How can we synchronize logical-sink?
- 8. Our Solution An entity registers as UCI in MediaSense Each logical-sink registers itself as a UCI and associated context-IDs as its data Logical-sinks are synchronized by registering physical sinks as UCIs
- 9. Our Approach
- 10. Our advantages Real-time, Distributed- no central point of failure issue Fast, Efficient, Scalable Memory efficient
- 11. Our advantages 74% 74% 74% improvement compared to existing MediaSense # of published context-IDs Current MediaSense Modified MediaSense % improvement 1000 7.34 ms 4.17 ms 76 10000 8.93 ms 5.37 ms 66 100000 10.74 ms 6.23 ms 72 200000 11.65 ms 6.69 ms 74
- 12. Analysis For both Published and Subscribed items 3537 messages/sec if it is run only for one second and over 9000 if just published for logical-sink PubSub messages/sec lowers by one-third while magnitude is increased by ten-fold
- 13. Analysis 86% Subscription matching only For a hundred-fold increase 86% increase in matching duration
- 14. Analysis Subscription matching only for a single context-ID The one-millionth context-ID took 8.76 ms to match with the published context-IDs
- 15. Analysis 99% 99% 99% improvement compared to PARDES for 2 million context-IDs PARDES increasing; MediaSense increase is minimal
- 16. Analysis (Scalability) # of context- IDs Le Subscribe (Counting) MediaSense % improvement 500 K 85 ms 14.76 ms 476 1 million 350 ms 16.22 ms 2058 # of context- IDs Le Subscribe (Counting) MediaSense % improvement 15 K 621 3151 407 1 million 7 91 1200 2058% 2058% improvement compared to Le Subscribe in terms of subscription matching 1200% 1200% improvement in PubSub messages/sec
- 17. Analysis 451% 163% 163% & 451% MediaSense betters Le Subscribe and ToPSS respectively by 163% and minimum by 451%
- 18. Conclusions MediaSense is feasible as PubSub model in crowdsourcing especially in logical-clustering It is fast- requires only 9.59 ms to match two-millionth published context-ID Scales well compared to other PubSub model Efficient and no centralization Occupies only 185.97 MB memory to store 5 million context-IDs
- 19. Future Work UCI discovery Performance evaluation on devices with limited computational capabilities Adaptability and awareness Security might be a concern; we will look into this
- 20. + 46 (0) 70 7463968 Contact Hasibur Rahman hasibur.rahman021 hasibur@dsv.su.se twitter.com/hasiburrahman29 facebook.com/SuZon.Hasibur.Rahman
- 21. ACKNOWLEDGMENT The work is partially supported by funding from the European Union FP7 MobiS project. The outcome of this research will be used later in the project.
- 22. Q & A! Thank You!
- 23. Credits mobiS logo www.mobis-euproject.eu/ IMAGE CREDITS Q & A image http://www.openlounge.org/lunargame/one-question-infinite- answers/