2015 State of the PhD, Barcelona, Spain

This presentation highlights the status of the PhD thesis as of June 2015. The talk was given at EMJD-DC Spring Event in Barcelona, Spain. For more details on the work, visit http://aminmkhan.com.

To address this limitation and enhance utility of community networks, we propose in this Phd thesis building a community cloud system that employs resources contributed by the members of the community network for provisioning infrastructure and software services. The design of such a community cloud needs to be tailored to the specific social, economic and technical characteristics of the community networks for it to be successful and sustainable. We focus in this thesis to explore the socio-economic mechanisms that can help in adoption and growth of community cloud model, and to fit them into a distributed architecture of cloud systems that is adapted to the scenarios of community networks. Our aim is that in the long term, cloud-based innovative applications can be shaped by developers from community cloud services and users of community networks will not not only consume applications from the Internet, but will also find them in clouds within community networks.

Transcript

1. Community Clouds in Community Networks: Economic Mechanisms Progress Update for Doctorate in Distributed Computing Amin Khan amin.khan@ieee.org / aminmkhan.com Advisors Felix Freitag Luís Rodrigues Universitat Politècnica de Catalunya Instituto Superior Técnico. INESC-ID Barcelona, Spain Lisbon, Portugal Monday 1 June 2015 EMJD-DC Spring Event, Barcelona, Spain
2. Introduction 2
3. Education European Masters in Informatics, EuMI (2005-2007) – Università degli Studi di Trento, Trento, Italy – University of Edinburgh, Edinburgh, UK BE in Computer Software Engineering (2001-2004) – National University of Sciences and Technology (NUST) – Rawalpindi, Pakistan 3
4. Erasmus Mundus Alumni Program Representative for 2015 – Erasmus Mundus Joint Doctorate in Distributed Computing Member of – Community Development Team – PhD Network – Mentors Initiative 4
5. Projects Past Research Projects – The Effects of Distribution and Parallelisation on the Performance of OGSA-DAI1 Enactments – Grid Services for Distributed Database Access for GAE project2 – Reputation Lending to counter free-riding in P2P Systems Past Enterprise Projects – Book Publishing Software System3 (US & Canada) – Retail Advertising Management System4 (US) 1http://www.ogsadai.org.uk 2http://pcbunn.cacr.caltech.edu/GAE/GAE.htm 3http://www.acumenbook.com/ 4http://knightware-inc.com/dp.htm 5
6. Interests Computational Linguistics Social Anthropology and Political Economy Cognitive Science and Human-Machine Interaction and Behavioural Economics 6
7. Research Problem 7
8. Community Clouds in Community Networks Community Networks – social collective for building ICT infrastructure – bottom-up connectivity based on reciprocal sharing Community Clouds – cloud services in community networks – based on voluntary and p2p computing models – federated, micro-, multi, and edge-clouds – serving local interests of community Community Clouds as much a social construct as a technical one! 8
9. Economic Mechanisms Research Questions – how, why and where can community clouds be useful? – how to speed-up uptake and ensure sustainability of community clouds? – how can we devise socio-economic mechanisms to sustain collaboration in community clouds? Economic Mechanisms – provide efficient and optimal resource allocation – build incentives to participate – ensure adherence to rules of the game 9
10. Past 10
11. Community Clouds State-of-the-art study of community clouds Decentralized architecture tailored to socio-economic context Comparison of distributed cloud models using CloudSim simulator 5 Socio-economic mechanisms for community clouds Decision support system for Community Clouds 6 5in collaboration with Leila Sharifi at IST Lisbon 6in collaboration with Victor Muntés at CA Labs Barcelona 11
12. Incentive-Based Resource Regulation Account for contribution when allocating resources More participation, more reward due to regulation Evaluation with simulations over federated community clouds Prototype implementation in CommunityLab testbed 7 Modelling provisioning and consuming virtual machines by independent players 7in collaboration with Umit Buyuksahin at UPC 12
13. Present 13
14. Auctions for Pricing Truthful and Trusted Auctions in an NP-Hard World Auctions useful for pricing where valuation is complex Numerous applications – Pricing arbitrarily-bundled VMs at public cloud providers – Trading VMs in federated cloud markets – Bandwidth reservation for VMs at public cloud providers – Pricing intra-data centre bandwidth for VMs – Network routing in multi-hop networks – Secondary Wireless Spectrum Auctions – Bandwidth reservation for cloud application in CNs Requirements for auctions – Cheating? Truthful, strategy-proof or incentive-compatible – Trust the service providers – Social welfare or Efficiency – Budgets – Performance 14
15. Truthful Auctions in an NP-Hard World Miracles of Randomized Algorithms Truthful and Efficient Auctions – Require optimal scheduling or allocation algorithms – Doesn’t work when finding optimal allocation is NP-Hard Randomized algorithms 8 – Guarantee polynomial running time in expectation – Guarantee truthful auction in expectation – Guarantee optimal allocation in expectation – Bounded by (1 − ϵ) times optimal solution 8Zhang, X. et al. (2015). A Truthful (1-ϵ)-Optimal Mechanism for On-demand Cloud Resource Provisioning. In INFOCOM 2015. 15
16. Trusted Auctions in an NP-Hard World Miracles of Cryptograph and Zero Knowledge Proofs Lack of trust in service provider Collusion between providers or buyers Guarantee perfect auction without disclosing anything to anyone 9 – Provider doesn’t know identity or actual bid values of any buyers – Buyer can claim anything about its value but can prove nothing about it! – Still provider proves to everyone ⋆ winners are chosen correctly ⋆ winners pay the right price Computationally too expensive to run in practice! 9Micali, S., & Rabin, M. O. (2014). Cryptography miracles, secure auctions, matching problem verification. Communications of the ACM, 57, 85–93. 16
17. Truthful and Trusted Auctions in Community Clouds Auction run by multiple nodes in parallel Distributed across the community network How to minimise the computation overhead? How to ensure correctness with a few Byzantine nodes? How to provide near-optimal allocation? 17
18. Work Plan 18
19. Quarterly Progress and Plan Year Qtr Description 2013 1st Background reading and state-of-art study 2nd Incentives for resource regulation 3rd Performance evaluation using CloudSim 4th Incentives for resource regulation 2014 1st Economic mechanisms & Distributed architecture 2nd Internship at CA Labs for MODAClouds 3rd Extend economic resource regulation work 4th Auction mechanisms for resource allocation 2015 1st Auction mechanisms for resource allocation 2nd Auction mechanisms for resource allocation 3rd Experiments using community cloud distribution 4th Thesis writing & PhD Pre-Defence 2016 1st Finalize thesis/papers & PhD Defence 19
20. Future 20
21. The Next Big Thing Edge clouds, Cloudlets, Nano data centres, Micro clouds Cloud providers and more, all have a stake in the pie 10 – Microsoft, Ericsson, HP, Akamai, Nokia, Motorola, IBM, etc. Triggers – Latency requirements for mobile applications – Computations on data in-transit – Network virtualisation Community clouds at the edge Bottom-up infrastructures complementing data centres Evolutionary game models for users participation 10Victor Bahl (2015). The Emergence of Micro Datacenters (Cloudlets) for Mobile Computing. Devices and Networking Summit 2015. Paris, France. 21
22. Thank You! @aminmkhan aminmkhan.com 22