Delay Tolerant Networking. Thomas Plagemann Distributed Multimedia Systems Group Department of Informatics University of Oslo.

Transcription

1 Delay Tolerant Networking Thomas Plagemann Distributed Multimedia Systems Group Department of Informatics University of Oslo Outline Background, motivation, overview Epidemic routing Message ferrying Mobility/density space Acknowledgement: Many transparencies are from Mustafar Ammar s keynote talk at Co-Next

2 Traditional Wired Networks endsystem (source) router endsystem (destination) separation between endsystems and routers routers responsible for finding stable path Traditional Mobile Ad-hoc Wireless Networks (MANET) node (source) node (destination) node = endsystem + router no separation between endsystems and routers nodes responsible for finding stable path 2

3 Traditional Mobile Ad-hoc Wireless Networks (MANET) node (source) node (destination) nodes may move routing layer responsible for reconstructing (repairing) stable paths when movement occurs The Traditional MANET Wireless Paradigm The Network is Connected There exists a (possibly multi-hop) path from any source to any destination The path exists for a long-enough period of time to allow meaningful communication If the path is disrupted it can be repaired in short order Looks like the Internet above the network layer 3

4 The Rise of Sparse Disconnected Networks Sparse Wireless Networks Disconnected By Necessity By Design (e.g. for power considerations) Mobile With enough mobility to allow for some connectivity over time Data paths may not exist at any one point in time but do exist over time 4

5 Mobility-Assisted Data Delivery: A New Communication Paradigm Mobility used for connectivity New Forwarding Paradigm Store Carry for a while forward Special nodes: Transport entities that are not sources or destinations Data Applications Nicely suitable for Message-Switching Delay tolerance but can work at multiple time scale (a.k.a. Delay Tolerant Networks ) 5

6 Some Delay-Tolerant Systems ZebraNet and SWIM Data MULE and Smart-Tags Vehicle-to-Vehicle Communication DakNet Epidemic Routing Message Ferrying SWIM 6

7 Vehicles on Highways Networks Destination Source Vehicles on Highways Networks Destination Source 7

8 Vehicles on Highways Networks Destination Source DakNet (Pentland, Fletcher, and Hasson) 8

9 Epidemic Routing Vahdat and Becker Utilize physical motion of devices to transport data Store-carry-forward paradigm Nodes buffer and carry data when disconnected Nodes exchange data when met data is replicated throughout the network Robust to disconnections Scalability and resource usage problems Epidemic Routing The Idea 9

10 Epidemic Routing The Idea Epidemic Routing The Idea 10

11 Epidemic Routing The Idea message is delivered Epidemic Routing Basic Elements Each node contains Message buffer Hash table Summary vector List of last seen nodes 11

12 Epidemic Routing The Protocol [Vahdat & Becker, TechReport 200] Epidemic Routing Multiple Hops Each message contains: Unique message ID Hop count Ack request (optional) Tradeoff buffer size vs. message delivery 12

13 Epidemic Routing Evaluation Implementation in ns-2 ERA ERA ERA ERA ERA Epidemic Routing Agent IMEP IMEP IMEP IMEP IMEP Internet MANET Encapsulation Protocol IEEE MAC protocol Model of radio propagation Model of node mobility 50 mobile nodes Area 1500m x 300m Random waypoint Speed 0 20 m/s (uniformly distributed) Message size 1 KB 45 message sources/sinks (each sends one message to the others) Each second 1 message Epidemic Routing Evaluation [Vahdat & Becker, TechReport 2000] 13

14 Epidemic Routing Evaluation [Vahdat & Becker, TechReport 200] Epidemic Routing Evaluation [Vahdat & Becker, TechReport 2000] 14

15 Epidemic Routing Evaluation [Vahdat & Becker, TechReport 2000] Epidemic Routing Evaluation [Vahdat & Becker, TechReport 2000] 15

16 The Trouble with ER Potentially high-failure rate Message duplication consumes nodal resources Some mobility patterns can cause disconnection Can be improved with contact probability information - Levine et al Message Ferrying GT Zhao and Ammar Exploit non-randomness in device movement to deliver data A set of nodes called ferries responsible for carrying data for all nodes in the network Store-carry-forward paradigm to accommodate disconnections Ferries act as a moving communication infrastructure for the network 16

17 Message Ferrying The Idea MF S M S MF M D D MF Variations Ferry Mobility Task-oriented, e.g., bus movement Messaging-oriented, e.g., robot movement Regular Node Mobility Stationary Mobile: task-oriented or messaging-oriented Number of ferries and level of coordination Level of regular node coordination Ferry designation Switching roles as ferry or regular node 17

18 MF for Networks with Mobile Nodes Nodes are mobile and limited in resources, e.g., buffer, energy Single ferry is used Not limited in buffer or energy Trajectory of the ferry is known to all nodes Data communication in messages Application layer data unit Message timeout Four Approaches Non-Proactive ( = Messaging-Specific) mobility Ferrying without Epidemic Routing Ferrying with Epidemic Routing Proactive Routing Schemes Node-Initiated MF(NIMF) Nodes move to meet ferry Ferry-Initiated MF (FIMF) Ferry moves to meet nodes 18

19 Node-Initiated Message Ferrying Meet the If no, keep OK working ferry? Working Node-Initiated Message Ferrying Go to Ferry 19

20 Node-Initiated Message Ferrying Send/Recv Go to Work Node-Initiated Message Ferrying Go to Work 20

21 Mode Transition WORKING GO TO FERRY Not planned Intentional GO TO WORK SEND/RECEIVE detour: whether the node is detouring; Node Operation in NIMF mode: which mode the node is in; 1. WORKING mode detour = FALSE; IF Trajectory Control indicates time to go to the ferry, detour = TRUE; mode = GO TO FERRY; On reception of a Hello message from the ferry: mode = SEND/RECV; 2. GO TO FERRY mode Calculate a shortest path to meet the ferry; Move toward the ferry; On reception of a Hello message from the ferry: mode = SEND/RECV; 3. SEND/RECV mode Exchange messages with the ferry; On finish of message exchange or the ferry is out of range: IF detour is TRUE, mode = GO TO WORK; ELSE mode = WORKING; 4. GO TO WORK mode Move back to node s location prior to the detour; On return to the prior location: mode = WORKING; On reception of a Hello message from the ferry: mode = SEND/RECV; [Zhao et al., MobiHoc04] 21

22 Ferry Operations in NIMF 1. Move according to a ferry route; 2. Broadcast Hello messages periodically; 3. On reception of an Echo message from a node: Exchange messages with the node; [Zhao et al., MobiHoc04] Node Trajectory Control Whether node should move to meet the ferry Goal: minimize message drops and reduce proactive movement Go to ferry if Work-time percentage > threshold and Estimated message drop percentage > threshold 22

23 Simulations Ns simulations using MAC and default energy model 40 nodes in 5km x 5km area 25 random (source, destination) pairs Node mobility random-waypoint with max speed 5m/s Message timeout: 8000 sec Single ferry with speed 15m/s Rectangle ferry route Performance Metrics Message delivery rate Message Delay Number of delivered messages per unit energy Only count transmission energy in regular nodes 23

24 Message Delivery Rate FIMF NIMF F w/er ER Mes s age delivery rate Epidemic Routing 0.2 E pidemic R outing (w/ ferry) NIMF 0.1 FIMF- NN FIMF- TA Node buffer size (messages) Message Delay FIMF NIMF F w/er ER Mes s age delay (s ec ) Epidemic Routing E pidemic R outing (w/ ferry) 500 NIMF FIMF- NN 0 FIMF- TA Node buffer size (messages) 24

25 Impact of Node Mobility Pattern Mobility Model Scheme Delivery Rate Delay (sec) Energy efficiency (KB/J) Random Waypoint Limited Random Waypoint NIMF FIMF ER(w/ ferry) ER NIMF FIMF ER(w/ ferry) ER Where Does MF Fit? Consider the space of wireless mobile networks Two Important Dimensions Relative Mobility Density 25

26 Some Terminology A Space Path: A multi-hop path where all the links are active at the same time A Space/Time Path: A multi-hop path that exists over time NOTE: S path is a special case of S/T path See Example A Space Paths Network 26

27 Example A No Path Network Example A Space Time Path 27

28 Example A Hybrid Network The Mobile Wireless Space High Relative Mobility Hybrid Environments Space/Time Paths Low Space Paths No (Space/Time) Paths High Node Density Low 28

29 Mapping Solutions to Space High Mobility MF is applicable for the entire space Hybrid Environments Space/Time Paths MF is necessary here Low Space Paths Traditional MANET solutions apply here No (Space/Time) Paths High Node Density Low MAC Layer Support for Delay Tolerant Video Transport in Disruptive MANETs Morten Lindeberg, Stein Kristiansen, Vera Goebel and omas Plagemann University of Oslo, Department of Informatics [mglindeb, steikr, goebel, 29

30 Background and Motivation Emergency and rescue scenarios Video streaming services can improve communication and operational efforts E.g., firemen wearing head-mounted cameras Infrastructure destroyed or not existing Solution: Mobile ad hoc networks with possible disruptions 59 Target Scenario Mimic realistic scenario Enforce store-carry-forwarding 500 m Sr I 1 Sr - Source node I x - Intermediate node(s) Cr x- Carrier node(s) CCC - Command Control Center I 2 Cr 1 Cr 2 CCC 500 m I 3 Location of the accident I 5 I 4 Command and control center Distance=1750 m 60 30

31 Use existing technology? Video encoding: H.264 significantly reduce bandwidth requirements Transport: TCP does not work well Network: IP + MANET routing: OLSR Wireless: IEEE ad hoc mode 61 Transport: Overlay Solution A delay tolerant overlay Store-carry-forward paradigm Dts-Overlay Inspired by MOMENTUM [1] Messages are sent hop-by-hop Route through carrier nodes when no route exist Dts-Overlay <OverlayMessage> Dts-Overlay Problem: packet loss Rejected packets + Retransmission queue status [1] Cabrero, S., Paneda, X.G., Plagemann, T., Goebel, V.: Overlay solution for multimedia data over sparse MANETs. In: International Wireless Communications and Mobile Computing Conference, IWCMC (2009) Route UDP availability UDP + MAC address IP / OLSR /ARP status IP / OLSR MAC MAC (IEEE a/b) (IEEE a/b) PHY 62 PHY (IEEE a/b) (IEEE a/b) Node 1 Node 2 31

32 Packet loss OLSR routes with broken links In essence, lower layers are not delay tolerant Address Resolution Protocol (ARP) 1. ARP will loose packets if no nodes reply with matching MAC address to the broadcasted IP address IEEE MAC Packets silently dropped after final retransmission 2... or if queue exceeds limit 63 Solution ARP Support: Avoid packet loss by only forwarding when address is resolved MAC Support (IEEE b): MAC Return: Do not drop packet after final retransmission, hand it to Dts-Overlay Link Adapt: Do not forward if the MAC transmission queue is filling (link is probably down) Reduce non-meaningful re-transmissions (MAC re-transmission limit) 64 32

33 Implementation Implemented in ns-3 (network simulator) Workload: Foreman CIF resolution (352x288) 12 second duration on repeat H.264 baseline profile Target bitrate 256 kb/s Wireless model: IEEE b ad hoc mode DSSS modulation Friis propagation loss model Constant speed propagation delay model 65 Evaluation Metrics: 1. Packet reception (Rxv) 2. Buffered packets (Buf) 3. Packet loss (Pl) 4. Sum of bytes (video) at PHY layer (Txtot) Present average from five different runs 3600 s duration Main-Studies (3600 s duration): ER (custom scenario) Sparse MANET (1250 m x 1250 m, 20 nodes, random walk) Dense MANET (250 m x 250, 20 nodes and additional workload, random walk) 66 33

34 Results: Pre-Study Address Resolution Affect packet loss substantially! We achieve lowest packet loss with Dynamic ARP!.. wait for ARP resolution before using a link Default Static Dynamic Pl 39.3 % 31.7 % 19.5 % 67 Configurations C3 C4 C5 C6 C7 Re-trans MAC Return No No No Yes Yes Link Adapt No No Yes No Yes 68 34

35 100 % Lowering retransmission limit cause packet loss 90 % 80 % 70 % 60 % 50 % 40 % 30 % 20 % 10 % 0 % Deploying Link Adapt does not affect packet reception much MAC Return close to eliminates all Results: packet loss ER C3 C4 C5 C6 C7 Link Adapts improves further, even reduces total transmission ov by 4% Pl Buf Rx 69 MAC Return improves reception Results: Sparse MANET Tendencies are the same, high standard deviation (mobility) Switch to single run Packet count (#) Sent C3 C4 C5 C6 C7 Tx t (MB) C3 C4 C5 C6 C7 With even less transmissions than default configuration Time (s) Time (s) 70 35

36 Results: Dense MANET Not much loss really, although some duplicates for MAC Return TxTot (MB) C3 C4 C5 C6 C7 Link Adapt has strong effect on transmissions TxTot (MB) 71 Conclusion MAC Support Evaluation: MAC Return drastically reduce packet loss Allow us to lowering MAC retransmission limit Link Adapt lowers overhead of transmissions that are not meaningful Usability is strengthened through evaluation also in a dense MANET 72 36

37 Summary Background, motivation, overview Epidemic routing Message ferrying Mobility/density space Acknowledgement: Many transparencies are from Mustafar Ammar s keynote talk at Co-Next