Wireless Mesh Networks Unplanned Community Mesh Networks A

Wireless Mesh Networks Unplanned Community Mesh Networks A. Zubow “Architecture and Evaluation of an Unplanned 802. 11 b Mesh Network”, J. Bicket, D. Aguayo, S. Biswas, R. Morris, Mobicom 2005

Introduction • Community wireless network – Share a few wired Internet connections – 2 approaches to constructing community networks: • Multi-hop network – – Nodes in chosen locations Directional antennas Requires well-coordination Great coverage • Access point – – 2 / 23 Clients directly connected Access points operate independently Do not require much coordination Small coverage

Introduction • Ambitious vision for community networks – Operate without extensive planning or central management – Provide wide coverage and acceptable performance • Design decisions in this paper – – Unconstrained node placement Omni-directional antennas Multi-hop routing Optimization of routing for throughput in a slowly changing network • Risks – Small radio ranges, low-quality links (TCP), interference (inside + outside of the network) • Purpose of this paper – Evaluate unplanned mesh architecture (Roofnet case study) – Describe Roofnet’s end-to-end characteristics 3 / 23

Introduction • What is Roofnet? – “Mesh networking" technology developed by MIT – Town-wide wireless network – Automatically calculates the best path and continuously monitors the network – Each Roofnet node is a small computer running Linux with an 802. 11 b card running in ad-hoc mode and a omni-directional antenna 4 / 23

Roofnet Design • Deployment – Over an area of about four square kilometers in Cambridge, MA – Most nodes are located in buildings • 3~4 story apartment buildings • 8 nodes are in taller buildings – Each Roofnet node is hosted by a volunteer user (no planning) • Hardware – PC, omni-directional antenna, hard drive … – 802. 11 b card • RTS/CTS disabled • Share the same 802. 11 b channel • Non-standard “pseudo-IBSS” mode – Similar to standard 802. 11 b IBSS (ad hoc) – Omit beacon and BSSID (network ID) 5 / 23

Roofnet Design • Software and Auto-Configuration – Linux, routing software (Click toolkit), DHCP server, web server … – Automatically solves these problems • Allocating addresses • Finding a gateway between Roofnet and the Internet • Choosing a good multi-hop route to that gateway – Addressing • • Roofnet carries IP packets inside its own header format and routing protocol Assigns addresses automatically Only meaningful inside Roofnet, not globally routable The address of Roofnet nodes – Low 24 bits are the low 24 bits of the node’s Ethernet address – High 8 bits are an unused class-A IP address block • The address of hosts – Allocate 192. 168. 1. x via DHCP and use NAT between the Ethernet and Roofnet 6 / 23

Roofnet Design • Software and Auto-Configuration (Cont’d) – Gateways and Internet Access • A small fraction of Roofnet users will share their wired Internet access links • Nodes which can reach the Internet – Advertise itself to Roofnet as an Internet gateway – Acts as a NAT for connection from Roofnet to the Internet • Other nodes – Select the gateway which has the best route metric • Roofnet currently has four Internet gateways 7 / 23

Roofnet Design • Routing Protocol – Srcr • Find the highest throughput route between any pair of Roofnet nodes • Source-routes data packets like DSR • Maintains a partial database of link metrics – Learning fresh link metrics • Forward a packet • Flood to find a route • Overhear queries and responses – Finding a route to a gateway • • • 8 / 23 Each Roofnet gateway periodically floods a dummy query When a node receives a new query, it adds the link metric information The node computes the best route The node re-broadcasts the query Send a notification to a failed packet’s source if the link condition is changed

Roofnet Design • Routing Metric – ETT (Estimated Transmission Time) metric • Srcr chooses routes with ETT • Predict the total amount of time it would take to send a data packet • Take into account link’s highest-throughput transmit bit-rate and delivery probability • Each Roofnet node sends periodic 1500 -byte broadcasts • Bit-rate Selection – 802. 11 b transmit bit-rates • 1, 2, 5. 5, 11 Mbits/s – Sample. Rate • Judge which bit-rate will provide the highest throughput • Base decisions on actual data transmission • Periodically sends a packet at some other bit-rate 9 / 23

Evaluation • Method – Multi-hop TCP • 15 second one-way bulk TCP transfer between each pair of Roofnet nodes – Single-hop TCP • The direct radio link between each pair of routes – Loss matrix • The loss rate between each pair of nodes using 1500 -byte broadcasts – Multi-hop density • TCP throughput between a fixed set of four nodes • Varying the number of Roofnet nodes that are participating in routing 10 / 23

Evaluation • Basic Performance (Multi-hop TCP) – The routes with low hop-count have much higher throughput – Multi-hop routes suffer from inter-hop collisions Theory (lossless links): 11 / 23

Evaluation • Basic Performance (Multi-hop TCP) – TCP throughput to each node from its chosen gateway – Round-trip latencies for 84 -byte ping packets to estimate interactive delay 12 / 23

Evaluation • Link Quality and Distance (Single-hop TCP, Multi-hop TCP) – Most available links are between 500 m and 1300 m and 500 kbits/s – Srcr • Use almost all of the links faster than 2 Mbits/s and ignore majority of the links which are slower than that • Fast short hops are the best policy 13 / 23 Throughput/distance of all available links (left); only links used in some route (right)

Evaluation • Link Quality and Distance (Multi-hop TCP, Loss matrix) – Median delivery probability is 0. 8 – 1/4 links have loss rates of 50% or more – 802. 11 detects the losses with its ACK mechanism and resends the packets – Links used by Srcr at the bit-rate chosen by Sample. Rate. 14 / 23

Evaluation • Effect of Density (Simulated from Single-hop TCP) – Mesh networks are only effective if the node density is sufficiently high – For each subset size n, a random set of n Roofnet nodes are selected. An estimate of the multi-hop throughput between every pair in the subset is computed, using only members of the subset as potential forwarders. – More than 1 kbytes/s 15 / 23

Evaluation • Effect of Density (Simulated from Single-hop TCP) – Network only starts to approach all-pairs connectivity when there are more than 20 nodes, corresponding to a density of about five nodes per square kilometer. – A denser network offers a wider choice of short highquality links though using them causes routes to have more hops 16 / 23

Evaluation • Mesh Robustness (Loss matrix, Multi-hop TCP) – The number of potentially useful neighbors each node has • Neighbor is defined as a node to which the delivery probability is 40% or more – The majority of nodes use many neighbors – Roofnet makes good use of the mesh architecture in ordinary routing 17 / 23

Evaluation • Mesh Robustness (Simulated from Single-hop TCP) – The extent to which the network is vulnerable to the loss of its most valuable links – The dozens of the best links must be eliminated before throughput is reduced by half 18 / 23

Evaluation • Mesh Robustness (Multi-hop TCP) – The y axis shows the average throughput among four particular nodes – The best-connected two nodes are important for performance – Losing both decreases the average throughput by 43% 19 / 23

Evaluation • Architectural Alternatives – Maximize the number of additional nodes with non-zero throughput to some gateway – Ties are broken by average throughput 20 / 23 Optimally chosen gateways Randomly chosen gateways

Evaluation • Comparison of the two tables shows that careful gateway choice increases throughput for both multi-hop and single-hop routing. • For five or fewer gateways, even randomly chosen multi-hop gateways provide better performance than carefully chosen single-hop gateways. • For larger numbers of gateways, however, carefully chosen single-hop gateways are better than randomly-chosen multi-hop gateways.

Evaluation • Inter-hop Interference (Multi-hop TCP, Single-hop TCP) – Concurrent transmissions on different hops of a route collide and cause packet loss Eq. 1: 22 / 23 Each point on the graph represents one node pair. The y-value of the point shows the measured throughput between that pair of nodes. The x-value shows the throughput predicted along that route by Equation 1 and the single-hop TCP data-set.

Network Use • Measurements of user activity – One of the four Roofnet gateways monitors the packets forwarded between Roofnet and the Internet – In one 24 -hour period • Average of 160 kbits/s between Roofnet and the Internet • Data was 94% • 48% of the data traffic was to or from nodes one hop form the gateway, 36% two hops • The gateway’s radio was busy for about 70% of the monitoring period • Almost all of the packets were TCP, less than 1% were UDP • 30% of the total data transferred was P 2 P file sharing program 23 / 23

Conclusions • The network’s architectures favors – Ease of deployment – Omni-directional antennas – Self-configuring software – Link-quality-aware multi-hop routing • Evaluation of network performance – Average throughput between nodes is 627 kbits/s – Well served by just a few gateways whose position is determined by convenience – Multi-hop mesh increases both connectivity and throughput 24 / 23

Resources • “Architecture and Evaluation of an Unplanned 802. 11 b Mesh Network”, J. Bicket, D. Aguayo, S. Biswas, R. Morris, Mobicom 2005