1. Data Dissemination for Spot Applications in Ad-Hoc Networks Alaeddine EL-FAWAL Thesis director : Jean-Yves Le Boudec Private Defense, 3 April 2009, EPFL

2. Overview on the PhD Work Supported by 2 projects: Haggle: European project in Situated and Autonomic Communications NCCR MICS: National Center of Competence in Research – Mobile Information and Communication Systems 2 parts in my PhD work: Data dissemination for spot applications over WIFI technology (Haggle) Cross-layer optimization for UWB systems (MICS) Dealing with different facets of uncoordinated ad-hoc wireless networks and deals with challenges at all networking layers

3. 3 OUTLINE Introduction Conclusions Prototyping and Testbed SLEF: Our Data Dissemination Middleware Performance Validation Achievements

4. 4 Open-Ended Environment Caused by: Dramatic expansion of WIFI interfaces: laptops, PDAs, mobile phones, video games and even with peripherals and vehicles. Characteristics:  Involves from few to thousands of nodes.  Challenging circumstances:  Highly dynamic  Unpredictable  Uncoordinated  Short contact time  Quickly changing from dense to sparse, non-congested to congested

5. 5  Destination: all nodes within the spot (multi-hop).  The spot might be the entire network (campus).  Example: ad applications, traffic info, support routing, resource discovery, bootstrapping phases for application layer. Spot Applications Description

6. 6  The spot size is variable according to the network conditions (trade-off spread-application rate)  Delivery ratio can not be used… instead, we talk about spread.  Efficient and reliable data dissemination service  Trade-off: spread-application rate Spread: number of nodes within the application spot (number of nodes that receive a packet). Trade-Off Spread vs. Application Rate N : Spread FF: Forwarding Factor R 0 : Rate of MAC Layer λ : Application Rate Requirements: Spot Applications

7. 7  Efficient and reliable data dissemination service  Trade-off: spread-application rate Spread: number of nodes within the application spot (number of nodes that receive a packet). The spot size is variable according to the network conditions.  Delivery ratio can not be used… instead, we talk about spread. Requirements: Spot Applications

8. 8 Open-Ended Envir. : Variation and Diversity of Scenarios Transmission range Density: average Few sources: little new injected traffic Relay Source / Relay

9. 9 Transmission range Density: average almost all are sources: a lot of new injected traffic Relay Source / Relay Open-Ended Envir. : Variation and Diversity of Scenarios

10. 10 Relay Source / Relay Very high density (traffic jam) almost all are sources: a huge amount of new injected traffic One hop: + 200 neighbors Open-Ended Envir. : Variation and Diversity of Scenarios

11. 11 Relay Source / Relay An autonomic mechanism for data dissemination that adapts to this diversity in scenarios is a must, otherwise network failure Density: very sparse. No communication without mobility (opportunistic communication) Open-Ended Envir. : Variation and Diversity of Scenarios

12. 12 OUTLINE Introduction Conclusions Prototyping and Testbed SLEF: Our Data Dissemination Middleware Performance Validation Achievements

13. 13 Data dissemination through limited epidemic forwarding : A source transmits packets in broadcast mode. Nodes forward each packet they receive Forwarding Factor times. Packets are forwarded within a limited hop count. SLEF: Self Limiting Epidemic Forwarding We propose the SLEF middleware. Delivers an efficient and reliable data dissemination service to the spot applications Deals with the tradeoff: spread-application rate The network conditions define the TTL limit (spread control). Functional between the application and the transmission sockets (UDP or raw sockets).

14. 14 SLEF: Self Limiting Epidemic Forwarding SLEF is designed to hold in all scenarios, in particular in very dense and very sparse ones SLEF Features:  Autonomic: Adapts itself to any change in the network. Density increases  forwarding factor decreases Traffic load increases  TTL limit decreases  Complete design of a middleware  Does not need / exchange any topology information. Uses only local information to the node (very short contact time).

15. 15 Essential Functions for Multi-Hop Broadcast SLEF implements 6 essential functions needed for a sustainable service 1. Congestion control: first mechanism proposed for broadcast in ad hoc networks 4. Spread control (adaptive TTL) 5. Forwarding factor control 6. Buffer management 2.Efficient use of MAC broadcast: 802.11 broadcast does not implement any exclusion mechanism (RTS/CTS) and it performs poorly (similar to Aloha). We replace it by a new scheme that we call pseudo-broadcast. 802.11 broadcast does not implement Ack. Pkts might be transmitted in the vacuum. We implement a presence indicator that does not need any message exchange. 3.Scheduler / fairness: A scheduler is needed to decide which packet to serve. It is based on Source ID to ensure some level of fairness.

16. 16 Spread Control: Adaptive TTL Trade-Off Spread vs. Application Rate Spread: number of nodes that receive a packet. N : Spread FF: Forwarding Factor R0: Nominal Rate of MAC Layer λ : Application Rate  We use an Aging mechanism  Adaptive TTL: Aging adapts locally the TTL to the different network setting, based on the send/receive events.  The idea is as follows: How: TTL limit= 2 Density =>TTL : In a traffic jam: TTL = 1 Density => TTL : In a very sparse network: TTL = 10 With SLEF Density => Spread Rate With fixed TTL

17. 17 Aging New created pkt: Age = 0 Pkt received for the first time: Age = 255 – TTL Age manipulated locally. when transmitting: TTL = 255 -Age Hop count: plays the role of a fixed TTL (=255/K0) if the network is not congested. Adaptive Age: adapts the TTL to the network activity, which reflects the density and the traffic load Real time Age: A pkt lives at most 8 hours (work cycle). Age Send/receive the same pkt Age = Age + K0 receive any pkt Age = Age + K1 Age > 255 Drop packet Constant increase by time: 8h -> 255 hop count adaptive AgeReal time Age

18. 18 Spread-Rate Balance SLEF maintains a spread-rate balance 2 parameters to adjust: K0, K1 Once Adjusted, they work well with all settings Adjusting the spread-rate balance according to the application needs Default values for K0 and K1 are computed in the thesis

19. 19 Why: To Minimize redundancy (save resources) Transmission would be redundant One send/receive event Two send/receive events: The green nodes are inhibited: smaller forwarding factor  Inhibit nodes from transmitting over sent/received pkts.  We compute a virtual rate for each packet based on the send/receive events.  Adaptive: the virtual rate Adapts locally the Forwarding Factor to the different network settings Forwarding Factor Control How:

20. 20 3- The pkt is allowed to be transmitted only after: current time + (similar to a back-off system) Virtual Rate The smaller the vRate is, the longer the back-off time is: The pkt might be dropped before being transmitted Virtual Rate: The max rate a packet is transmitted with R0 : nominal MAC rate [pkts/s] RcvCount : number of times the pkt is received SendCount : number of times the pkt is sent a and b : are coefficient less than 1: a=0.1, b=0.01 For each send / receive event on a given pkt: 1- The virtual rate is computed as follows: Unlike other mechanisms, the virtual rate based forwarding factor control allows transmitting the pkt multiple times if needed. 2- vRate decreases exponentially with send/receive events.

21. 21 Buffer Management  Cleans the buffer in order to keep space for new incoming packets.  Based on Aging.  Drops packets with the highest age Highest ageLived the highest number of send/receive events

22. 22 OUTLINE Introduction Conclusions Prototyping and Testbed SLEF: Our Data Dissemination Middleware Performance Validation Achievements

23. 23 Network Simulator: JIST-SWANS, A JAVA simulator for Ad Hoc networks MAC: 802.11/b Channel : Fading Setting Scenarios: Vehicular networks. Different network settings: node density, traffic load… Vehicular Mobility Simulator: STRAW, an extension of JIST-SWANS. It provides a mobility model based on the operation of the real vehicular traffic. Topo: 2-lanes road, speed limit 80Km/h Range: 300 m in average K0 = 25 K1 = 0.1

24. 24 Adaptation of the Spread to the Rate Rate Spread Adaptive TTL Density: 12 vehicles/Km

25. 25 Adaptation of the Forwarding Factor to the Density Very dense (Traffic jam) +200 neighbors Very sparse (Death Valley)

26. 26 Importance of the Pseudo-Broadcast Very dense (Traffic jam): +200 neighbors Idea: implement a mutual exclusion mechanism for broadcast in order to avoid collision

27. 27 OUTLINE Introduction Conclusions Prototyping and Testbed SLEF: Our Data Dissemination Middleware Performance Validation Achievements

28. 28 SLEF Prototyping 2 architectures: MAC Spot application SLEF Spot application SLEF UDP IP MAC + It is independent of the IP address + Practical in the absence of a centralized coordination where assigning IP @ is challenging – Complex with Windows – IP networking needs to be initialized + Straightforward with all platforms Raw sockets UDP sockets

29. 29 SLEF Prototyping 4 platforms: Windows XP: J2SE, C++ Windows Mobile: J2ME, C++ Linux: J2SE, C++ OpenWrt (Linux-like firmware for embedded system): C++ Resource-constrained devices:  Smartphone HTC S620: Windows Mobile 64MB RAM, 128 ROM 201 MHz  ASUS WL-500 GP wireless router OpenWrt 32MB RAM, 8MB Flash 266MHz SLEF is practical and performs well on very resource-limited devices

30. 30 Testbed Stress test: More than 50 devices communicating with each other for long time. Performance evaluation of SLEF through measurements Testbed features: Wireless router: ASUS WL-500GP Technical specifications: 8MB Flash, 32MB RAM, 266MHz, 2 USB 2.0 ports Firmware: OpenWrt Configurable wireless interface: using Atheros card with MadWIFI driver (setting RTS/CTS Th, Tx power, promiscuous mode, monitor mode, Tx queue length…). Mobility: using plumb batteries, +4 hrs lifetime with full power transmission at full rate Robustness.

31. 31 Measurement Design Nodes are distributed over 12 buildings in EPFL

32. 32 Measurement Design  Application: injects packets at fixed rate.  Application rate: can be reduced by the congestion control mechanism.  Comparison: SLEF vs. fixed-TTL  Buffer size of fixed-TTL: Small buffer size: We ran SLEF and we use the average buffer occupation obtained (620 packets). Large buffer size: 10 000 packets.  Fixed-TTL: Implements all functions of SLEF, otherwise it is not functional. Spread control is replaced by TTL (decremented by one for each hop) TTL limit is fixed Buffer management is based on the TTL and not on the age: packets with the smallest TTL are dropped first when the buffer is full Fixed-TTL limits the spread through 2 parameters: TTL limit and the buffer size.

33. 33 Measurement Results With fixed TTL:  Setting the buffer size per scenario is needed: Large buffer size  radundancy and low rate Small buffer size  small spread even in non-congested network  TTL based buffer management does not perform well. Higher redundancy Lower rate Smaller spread

34. 34 Measurement Conclusions SLEF Fixed-TTL 2 parameters to adjust: K0, K1 Once adjusted, they work well with all settings Aging based buffer management performs well Max Buffer size is 255/K1 (little formula) 2 parameters to adjust: MaxTTL and buffer size Needs to adjust whenever the network setting (density, traffic load, mobility,…) changes TTL based buffer management performs poorly.

35. 35 OUTLINE Introduction Conclusions Prototyping and Testbed SLEF: Our Data Dissemination Middleware Performance Validation Achievements

36. 36 Conclusions Autonomic: Adapts itself to any change in the network. Complete middleware Does not need/exchange any topology information. Uses only local information to the node. Works even in extreme scenarios (very dense/sparse…) Performs well on resource constrained devices We propose SLEF: data dissemination middleware for spot applications Prototyping SLEF for 4 platforms Building a testbed for wireless networks protocols SLEF performs better than fixed-TTL: Fixed-TTL is not adaptive TTL-based bufer management performs poorly

37. 37 Perspectives Collaboration with LCA3: Prof. Patrick Thiran and Adel Aziz Extensive series of measurements for wireless network protocols Performance evaluation of SLEF: Factorial analysis applied on measurement results. Measurements will consider varying scenarios: mobility, intermittent connectivity, different traffic load and density. Comparison with different variants of data dissemination protocols

38. 38 OUTLINE Introduction Conclusions Prototyping and Testbed SLEF: Our Data Dissemination Middleware Performance Validation Achievements

39. 39 Achievements  Multi-Hop Broadcast Middleware (SLEF): One conference paper.  Vulnerabilities in Epidemic Forwarding: One conference paper.  SLEF Prototyping and Experimental Testbed. Data Dissemination for Spot Applications (Haggle)  Robust Signal Acquisition in UWB Ad Hoc Networks: One journal paper, one conference paper and one patent (adopted by MICS).  Sleeping Mode for UWB IR Ad-hoc Networks: One journal paper. Cross-Layer Optimization for UWB IR Networks (MICS)

40. 40 Robust Signal Acquisition in UWB Ad Hoc Networks: Problem: Conventional detection method  assume power control, otherwise it fails.  Power control impractical in the absence of a centralized coordination (CDMA). S1 D2S2 D1 Near-far scenario Solution: Power independent detection method. 10 users, LOS Indoor Office Channel model by IEEE P802.15.4a Proba of Misdetection: P MD Total Error: E t =P MD + P FA P MD Performance Evaluation:

41. 41 Issues Addressed for the First Time  Spot Applications (introduced for the first time).  Trade-off: Spread – Application rate.  Spread control.  Congestion control in ad-hoc networks in broadcast mode.  Fairness with epidemic forwarding.  Identifying vulnerabilities that are specific to epidemic forwarding.  First prototype of network coding for ad-hoc networks.  Power independent signal acquisition for UWB in uncoordinated wireless ad- hoc networks (problem identification and solution).  Identifying key design elements for power saving with UWB IR systems.

42. 42 Used Expertise  Queuing theory  Performance evaluation tools  Wireless networks  Security  All layers of TCP/IP stack, mainly: MAC (Medium Access Control) Physical Layer (UWB IR and 802.11)  Vehicular networks  Network coding  Different channel models  Signal processing.  Simulation ns-2, Jist-Swans, Straw (vehicular traffic simulator), Matlab.  System programming Windows XP, Windows Mobile, Linux, OpenWrt C++, J2SE, J2ME, raw sockets

43. 43 Demo: Ad-Hoc Ventes Flash Spot application Needs SLEF Windows platform: XP and Mobile C++ Sender (shop) Client Injects ads:  Shop name  Product features: Name Category Price … Receive ads:  Filtering at the application level: Shop name Category  Vote

44. 44 Demo: Ad-Hoc Ventes Flash Scenario: Persistent broadcast in presence of intermittent connectivity ShopClient 1Client 2 Application on X X X Application off X X X Application on Application off

45. 45 List of Publications at EPFL El Fawal, AlaeddineEl Fawal, Alaeddine ; Le Boudec, Jean-Yves: A Robust Signal Detection Method for Ultra Wide Band (UWB) Networks with Uncontrolled Interference. In: IEEE Transactions on Microwave Theory and Techniques (MTT,) 2006.Le Boudec, Jean-Yves Raya, MaximRaya, Maxim ; Aad, Imad ; Hubaux, Jean-Pierre ; El Fawal, Alaeddine DOMINO: Detecting MAC layer greedy behavior in IEEE 802.11 hotspots. In: IEEE Transactions on Mobile Computing, December 2006Aad, ImadHubaux, Jean-PierreEl Fawal, Alaeddine El Fawal, Alaeddine ; Le Boudec, Jean-Yves ; Salamatian, Kave Multi-hop Broadcast from Theory to Reality: Practical Design for Ad Hoc Networks. In: First International Conference on Autonomic Computing and Communication Systems, October 2007.Le Boudec, Jean-YvesSalamatian, Kave El Fawal, AlaeddineEl Fawal, Alaeddine ; Le Boudec, Jean-Yves ; Salamatian, Kave Vulnerabilities in Epidemic Forwarding In: The First IEEE WoWMoM Workshop on Autonomic and Opportunistic Communications (AOC2007), 2007Le Boudec, Jean-YvesSalamatian, Kave Merz, RubenMerz, Ruben ; El Fawal, Alaeddine ; Le Boudec, Jean-Yves ; Radunovic, Bozidar et al. The Optimal MAC Layer for Low-Power UWB is Non-Coordinated. In: IEEE International Symposium on Circuits and Systems (ISCAS 2006), 2006El Fawal, AlaeddineLe Boudec, Jean-YvesRadunovic, Bozidar El Fawal, AlaeddineEl Fawal, Alaeddine ; Le Boudec, Jean-Yves A Power Independent Detection Method for UltraWide Band (UWB) Impulse Radio Networks. In: IEEE International Conference on Ultra-Wideband (ICU 2005), 2005Le Boudec, Jean-Yves El Fawal, AlaeddineEl Fawal, Alaeddine ; Le Boudec, Jean-Yves Le Boudec, Jean-Yves Synchronizing Method for Impulse Radio Networks, Date: 2005 El Fawal, AlaeddineEl Fawal, Alaeddine ; Le Boudec, Jean-Yves et al. Tradeoff Analysis of PHY-aware MAC in Low-Rate, Low-Power UWB networks. In: IEEE Communications Magazine, vol. 43, num. 12, 2005, p. 147.Le Boudec, Jean-Yves Journal Articles: Proceedings: Pending Patent: El Fawal, AlaeddineEl Fawal, Alaeddine ; Salamatian, Kave et al. A framework for network coding in challenged wireless network. Presented at: MobiSys 2006, Uppsala - Sweden, June 19-22, 2006.Salamatian, Kave Demo

46. 46

47. 47 Efficient and Reliable Broadcast Presence indicator: (reliability) To avoid transmitting in the vacuum. A transmission is considered as successful if reception occurs a few seconds before or later, otherwise the transmission is repeated. Pseudo-broadcast: (efficiency) To avoid collision. Nodes transmit in unicast mode using RTS/CTS. The destination is the source of the last received packet. All neighbors, listening in promiscuous mode, receive the packet.

48. 48 Congestion Control  At most σ packets in the epidemic buffer.  An application can inject a new packet if: Either less than σ self packets in the epidemic buffer Or one of the self packets can be removed  A self packet can be removed if: Either transmitted at least once and Age > MaxAge Or transmitted at least once with implicit Ack (heard forwarded by neighbors)

49. 49 Testbed Testbed features: Wireless router: ASUS WL-500GP Technical specifications: 8MB Flash, 32MB RAM, 266MHz, 2 USB 2.0 ports Firmware: OpenWrt Configurable wireless interface: using Atheros card with MadWIFI driver (setting RTS/CTS Th, Tx power, Tx queue length…). Mobility: using plumb batteries, +4 hrs lifetime with full power transmission at full rate Robustness.