1 / 62

Wireless Mesh Networks

Wireless Mesh Networks. Paul A.S. Ward Shoshin Distributed Systems Group University of Waterloo pasward@ccng.uwaterloo.ca. Outline. Introduction Modeling and Capacity Multi-radio and Multi-channel Systems Congestion Control, Fairness, Load Balancing and Quality of Service

jewell
Télécharger la présentation

Wireless Mesh Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Wireless Mesh Networks Paul A.S. Ward Shoshin Distributed Systems Group University of Waterloo pasward@ccng.uwaterloo.ca

  2. Outline • Introduction • Modeling and Capacity • Multi-radio and Multi-channel Systems • Congestion Control, Fairness, Load Balancing and Quality of Service • Deployment Issues Wireless Mesh Networks: CNSR 2006 Tutorial

  3. What? A multi-hop wireless network, typically with • limited/zero mobility • powered • acting as an access network Other names include: • infrastructured ad hoc network • community area network Wireless Mesh Networks: CNSR 2006 Tutorial

  4. Variations: 802.11 WLAN DS • The distribution system of a WLAN using wireless forwarding AP Distribution System AP Gateway Internet AP Wireless Mesh Networks: CNSR 2006 Tutorial

  5. Internet Community-Area Networks J. Jun, M. L. Sichitiu, Department of Electrical and Computer Engineering, North Carolina State University Wireless Mesh Networks: CNSR 2006 Tutorial

  6. Residential Broadband Source: Broadband & Dial-Up Access Source: Leitchman Research Group Source: Victor Bahl, MRFS 2005 Wireless Mesh Networks: CNSR 2006 Tutorial

  7. Processing One PC-day of computation Storage 1 GB disk storage(2 DVD quality movies) Interconnection 100 MB broadband data (3.5 hours of music) 1 MB voice telephony(15 minutes talk time) 1.6 KB SMS (10 messages) Bits ≠ Value Broadband: 1¢ per MB GPRS: $1 per MB SMS: $600 per MB What can you get for a $1? It’s the Bandwidth (and Spectrum) that’s expensive Wireless Mesh Networks: CNSR 2006 Tutorial Source: Victor Bahl, MRFS 2005

  8. What not wire the last mile? The Last Mile: Connection between a home and local hub Scale & legacy make last mile expensive • ~ 135 million housing units in the US (U.S. Census Bureau 2001) • POTS (legacy) network designed for voice & built over 60 years • Cable TV networks built over last 25 years The Truck Roll Problem: Touching each home incurs cost: customer equipment; installation & servicing; and central office equipment improvements • In our estimate building an alternate, physical last mile replacement to hit 80% of US homes will take 19 years and cost ~ US $60-150 billion Source: Victor Bahl, MRFS 2005 Wireless Mesh Networks: CNSR 2006 Tutorial

  9. Relevant Deployments • Packet Radio – 1970s • Ricochet – mid- to late 90s • Roofnet – post 2000 • 802.11 cards w. Soekris single-board PCs • Nortel: Taipei: 10,000 nodes • WLAN replacement • With multi-channel, single radio 802.11a backhaul • Max length: two hops Wireless Mesh Networks: CNSR 2006 Tutorial

  10. Key Properties • Wireless • Multi-hop • Contrast with WLAN • Stationary • Powered • Gateway • Single Authority Domain (?) • Contrast 3-6 with ad hoc networks Wireless Mesh Networks: CNSR 2006 Tutorial

  11. Modeling and Capacity • Modeling Wireless Communication • Modeling Multi-hop Networks • Capacity Calculations Wireless Mesh Networks: CNSR 2006 Tutorial

  12. CSMA/CA MAC ~ 802.11 if (carrier(busy)) BEB; else { wait DIFS ; if (carrier(busy)) BEB; else { transmit(msg); BEB; } } Wireless Mesh Networks: CNSR 2006 Tutorial

  13. CSMA/CA MAC ~ 802.11 BEB() { WT = random(0 to CW) wait(WT); } Try again: double CW on each attempt, up to 1023 • Freeze countdown when medium busy • RTS/CTS may be used to avoid collisions Wireless Mesh Networks: CNSR 2006 Tutorial

  14. Single Node Communication (1) 2 • Locality of transmission • can communicate only within transmission range • Broadcast medium • Transmission range is “Tx” • Reality Check: • Reception range is not a uniform circle • Packet loss occurs due to signal degradation, especially caused by multi-path fading 4 1 3 Wireless Mesh Networks: CNSR 2006 Tutorial

  15. Single Node Communication (2) • Interference: multiple transmissions within neighbourhood of the receiver will cause a collision at the receiver • Collision is not observed at the sender • Interference range is “If” • Reality Check: • Interference is a function of SNR • Carrier sensing is often assumed to be equivalent to interference range 4 2 3 1 Wireless Mesh Networks: CNSR 2006 Tutorial

  16. Multi-hop Model • Model as graph: • Vertex is computer node with wireless card • Edge between two vertices if the nodes can communicate • Reality check: what does “communicate” mean? • Is this undirected? Wireless Mesh Networks: CNSR 2006 Tutorial

  17. Capacity • Constraint satisfaction problem on network graph • Depends on traffic • which depends on routing • activity • link-resource constraints • Link contention • which depends on node location • traffic direction • medium-resource constraints Wireless Mesh Networks: CNSR 2006 Tutorial

  18. G1 G2 G3 G4 1 2 3 4 GW Network Feasibility: Streams • Streams – link-resource constraints G1 <= L12 G1+G2 <= L23 G1+G2+G3 <= L34 G1+G2+G3+G4 <= L4G Gi >= 0 • Depend on routing and activity Wireless Mesh Networks: CNSR 2006 Tutorial

  19. G1 G2 G3 G4 1 2 3 4 GW Network Feasibility: Links contend? • Links – medium resource constrains • Collision domains • Over-estimates contention • Cliques 23 CA CB 12 4G contends 34 L12+L23+L34+L4G <= C23 L12+L23+L34 <= CA L23+L34+L4G <= CB • Depends on link-contention: • bi-directional link assumption is common Wireless Mesh Networks: CNSR 2006 Tutorial

  20. Link Contention • Link-contention model options • Measure it • Endpoint of one link within interference range of endpoint of other link • bi-directional assumption • 2-hop contention: Endpoint of one link is within transmission range of the endpoint of the other • Under-estimates contention Wireless Mesh Networks: CNSR 2006 Tutorial

  21. Link Contention • Reality Check: • Tx < If • interference is not all or nothing • a node cannot identify its entire contention neighbourhood • realistic vs. omniscient neighbourhood information Wireless Mesh Networks: CNSR 2006 Tutorial

  22. The Computational Problem • AG <= C • A – usage matrix • G – stream throughput vector • C – medium capacity vector (bandwidth) • Use the actual single-hop medium bandwidth • The problem (for absolute fairness): find Gmax = max G : For all active i, Gi = G Wireless Mesh Networks: CNSR 2006 Tutorial

  23. 2G G Example: Clique Model B G G G G G G 3G 4G 5G 6G Gmax = B/18 Wireless Mesh Networks: CNSR 2006 Tutorial

  24. B 2G G 21G Example: CD Model G G G G G G 3G 4G 5G 6G 6G 10G 15G 18G 15G Gmax = B/21 Wireless Mesh Networks: CNSR 2006 Tutorial

  25. Accuracy of Computation Wireless Mesh Networks: CNSR 2006 Tutorial

  26. Accuracy of Computation Wireless Mesh Networks: CNSR 2006 Tutorial

  27. Accuracy of Computation Wireless Mesh Networks: CNSR 2006 Tutorial

  28. Statistical Data • Compute fair-share point on hydra graph • Compare to calculated capacity • Clique model under-estimates by 0.07%, on average • Collision domain model under-estimates by 2.3% on average • Deviation in ~10% in both cases • Provided RTS/CTS is not used Wireless Mesh Networks: CNSR 2006 Tutorial

  29. Why RTS/CTS is Bad in WMN Wireless Mesh Networks: CNSR 2006 Tutorial

  30. Summary • Two capacity models: • Clique • Collision domain • Clique is more-accurate (single channel) • But requires information that may not generally be available • Collision domain approach is very close to accurate • Within the deviation of the two models • Computed with readily available information • Neither model is accurate in the presence of RTS/CTS Wireless Mesh Networks: CNSR 2006 Tutorial

  31. Multi-channel and Multi-radio Wireless Mesh Networks: CNSR 2006 Tutorial

  32. Design Issues • Location-dependent contentionA transmission interferes with both sender’s and receiver’s neighbours • Spatial channel reuseNon-contending flows can proceed simultaneously • Distributed coordination • Hidden terminal problem • Relayed trafficInter- and intra-stream contention over links Wireless Mesh Networks: CNSR 2006 Tutorial

  33. Capacity Problems • Efficient use suggests multi-channel routing Wireless Mesh Networks: CNSR 2006 Tutorial

  34. Approaches • Multi-channel, one radio • Cheaper • Switching delay • Same (or worse) delay as single-channel in a given hop • In, then out separately • But better over multiple hops Wireless Mesh Networks: CNSR 2006 Tutorial

  35. Approaches • Multi-channel, multi-radio • More radios: more expensive • But still relatively cheap • But they interfere with each other • Use one in 2.4 GHz and one in 5 GHz band • Access vs. backhaul separation • e.g. Nortel approach • Multi-radio backhaul Wireless Mesh Networks: CNSR 2006 Tutorial

  36. Multi-radio capacity • Generalize capacity models by creating N sub-graphs, one per channel, and then using the same basic approach on each sub-graph • Clique performs poorly in three-channel, two interface case • Collision Domain is largely accurate Wireless Mesh Networks: CNSR 2006 Tutorial

  37. Analysis: multi-radio backhaul • How much gain do I need? • Factor of “C” or just use one channel that spreads over the range of the “C” channels • Any gain, since the channels are already there • Performance gains of more than a factor of three can be realized • two-interface, three-channel model • But only with careful channel allocation • Delay is substantially reduced because of parallel operation Wireless Mesh Networks: CNSR 2006 Tutorial

  38. Congestion Control, Fairness, QoS • Congestion Problem • Causes • Tradeoff on fairness and capacity Wireless Mesh Networks: CNSR 2006 Tutorial

  39. G G GW 1 2 S1 S2 S1 S1, S2 S1, S2 S1 S1, S2 S2 G G The Problem Wireless Mesh Networks: CNSR 2006 Tutorial

  40. Solutions • Separate traffic into different queues • Some benefit • But MAC solutions do not scale to network fairness • AQM solutions • But the bottleneck is not at the gateway, and shifts • Rate limit at source • Only known method that works • How do we tell the sources what to limit to? Wireless Mesh Networks: CNSR 2006 Tutorial

  41. Solutions: Distributed • Gather the information • Required for the computation of the fair-share • Compute the fair-share rate • According to the network model • Enforce the computed rate at stream origins • Leaky-bucket style Wireless Mesh Networks: CNSR 2006 Tutorial

  42. Stream Activity • Don’t want to waste bandwidth for inactive (temporary non-existent) streams • Additional constraints: For all inactive k, Gk = 0 • For now: two-valued activity information • It is dynamic=> a method to gather the information is necessary Wireless Mesh Networks: CNSR 2006 Tutorial

  43. Fair-share Computation • Feasibility assumption: • Feasible requests => WYWIWYG • We can compute the fair-share rates from: • a network feasibility model + • a fairness criterion Wireless Mesh Networks: CNSR 2006 Tutorial

  44. Stream Activity Processing • Who computes what, and when? • Our choice: • All nodes compute Gmax on their own • The stream activity information is distributed as soon as it changes Wireless Mesh Networks: CNSR 2006 Tutorial

  45. Stream Activity Distribution • Keep it simple • There are up to 2*N streams in a one-gateway WMN • Activity status of all (possible) streams can be piggybacked with DATA packets • Requires no change of IEEE 802.11 MAC Wireless Mesh Networks: CNSR 2006 Tutorial

  46. Piggybacking Issues (1) • Version control rule is necessary • Our rule: • A node hop-wise closer to the origin is more up-to-date on the activity of its streams • Requires symmetric spanning tree routing Wireless Mesh Networks: CNSR 2006 Tutorial

  47. 1 2 3 4 GW Piggybacking Issues (2) • Carrier packets are required • The “one-way” problem • Alleviate it with promiscuous mode • The stream deactivation problem • Monitor the upstream links from children • The silent subtree problem • When information is unavailable, assume thestreamsare active Wireless Mesh Networks: CNSR 2006 Tutorial

  48. Experiments • CBR over UDP / FTP over TCP • Fixed shortest-path routing • No RTS/CTS • Two link-contention models: • o+cl = • omniscient neighbour info (interference range) • cliques • r+cd = • realistic neighbour info (transmission range) • collision domains Wireless Mesh Networks: CNSR 2006 Tutorial

  49. Chain Experiment (1) Simulation result with Plain TCP Wireless Mesh Networks: CNSR 2006 Tutorial

  50. Chain Experiment (2) Simulation result (TCP) with our fairness rate-control algorithm Wireless Mesh Networks: CNSR 2006 Tutorial

More Related