Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks

Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks

Girish Nandagudi

Acknowledgements

This presentation is based on the paper “Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks” by Richard Draves, Jitendra Padhye and Brian Zill

Introduction

Routing in ad-hoc wireless networks has been an active area of research

Research is mainly motivated by mobile applications in battlefield and other ad-hoc networks

It is important to provide scalable routing in such environments, where mobile nodes dominate the network

Note: The image has been borrowed from an internet article in the website http://www.sensorsmag.com

Point of interest…

The aim is to improve the network capacity or the performance of individual transfers [by means of an efficient routing algorithm]

Challenge– To cope up with the problem of reduction in total capacity of the

network due to interference between multiple simultaneous transmissions

Possible Solution– Provide two radios per node, enabling the node to transmit and

receive simultaneously– Having two (or more) radios can improve robustness, connectivity

and performance– Advantage is that the nodes can utilize more of the radio spectrum

Other alternative solutions

Using directional antennas Improved MACs Channel switching

Diagnosing the multiple radio scenario

When the nodes in the network has multiple radios, the shortest path algorithm does not perform optimally

Given a choice between 802.11a and an 802.11b radio, the shortest path algorithm chooses the slower 802.11b radio since it has longer range

A shortest path algorithm that selects the path without ensuring that the hops are on different channels will almost certainly, does not perform well

Why a new routing metric?

Shortest-path routing has several drawbacks when it comes to routing in multi-hop wireless networks

ETX (expected transmission count) metric performs well in single-radio environment, but it does not perform well in environments having different data rates and multiple radios

ETX

ETX uses the underlying packet loss probability, both forward and reverse, denoted by pf and pr respectively to measure the expected number of transmissions including re-transmissions

ETX is denoted by:

ETX = Σ k * s(k) = 1

1 - pK = 1

∞

• The path metric is the sum of ETX values for each link in the path. Thereafter, the routing protocol selects the path that has the minimum path metric

Disadvantages of ETX

When we have two radios per node, one radio with an 802.11a and the other with 802.11b, ETX will transmit the data over 802.11b

ETX only considers the loss rates over the links, but not their bandwidths

ETX prefers to transmit over shorter paths, but not on longer paths in order to minimize global resource usage

ETX does not give preference to diverse-channel paths. Hence, it does not perform well in a scenario where two 802.11b radios are used

The MR-LQSR protocol

New metric, WCETT (Weighted Cumulative Expected Transmission Time) introduced

LQSR is a source-routed link-state protocol derived from DSR

Differences between DSR and the MR-LQSR protocol

DSR MR-LQSR

DSR assigns equal weight to all the links in the network. The path metric is simply the sum of link weights along the path.

MR-LQSR assigns weight depending on the transmission latency, bandwidth and the channel diversity of the link.

DSR implements shortest path routing. MR-LQSR uses the WCETT metric for routing.

MR-LQSR: Assumptions

All nodes in the network are stationary Each node is equipped with one or more 802.11

radio. These can be among 802.11a, 802.11b and 802.11g radios or a mixture of them.

The number of radios per node may not always be the same

If a node is equipped with one or more radios, they are tuned to different, non-interfering channels

MR-LQSR: Design Goals

The protocol should take both loss rate and bandwidth of a link into account while considering it for inclusion in the path

The path metric should be increasing. That is, if an hop is added to the existing path, the cost of the path should never decrease

The path metric should account for the reduction in throughput due to interference among links that operate on the same channel

The protocol assigns a weight to each link that is equal to the expected amount of time it would take to successfully transmit a packet of some fixed size S

This time depends on the link bandwidth and loss rate Now, the ETT of a link i between x and y nodes is denoted by

ETTi

Using the above notation, the WCETT can be derived as:

Computing path metric

WCETT = Σ ETTii = 1

n

Computing path metric II

It is desirable for the WCETT to consider the impact of channel diversity

In a two-hop path, if the hops are interfering, then the effective bandwidth of the channel is reduced to half due to the fact that only one hop can operate at a time

The assumption that the hops that are nearby and in the same channel always interfere holds almost true for short paths, but it might be somewhat pessimistic for longer paths

Computing path metric III

Assuming a n hop path and that the system has a total of k channels, we define Xj as:

Xj = Σ ETTi 1≤j ≤kHop i is on channel j

WCETT is taken as max(Xj)

Computing path metric IV

The metric, WCETT = max(Xj) favors paths along diverse channels

This metric achieves the third design goal, but not the second design goal

To achieve both the design goals, we can combine the two equations as follows:

WCETT = (1 – β) * Σ ETTi + β * max Xji = 1

n

1≤j ≤k

Interpreting the expression

Two possible ways:1. The first term reflects the sum of the

transmission times along all hops in the network. The second term reflects the set of all hops that will have the most impact on the throughput of this path.

2. We can view the equation as a tradeoff between throughput and delay.

Measuring ETT

ETT is defined as bandwidth-adjusted ETX Hence, ETT is given by

– ETT = ETX * (S / B)

To accurately calculate the ETT, we need to know the forward and reverse loss rates (pf and pr) and the bandwidth of each link

This can be achieved by using broadcast packet technique described by De Couto et al [2]

Measuring ETT - Determining bandwidth

Determining bandwidth is complex One possibility is to set the bandwidth of

each 802.11 radio to a fixed value Another possibility is to allow 802.11 radios

to select the bandwidth automatically by enabling them to operate at autorate mode

Measuring ETT - Determining bandwidth II

The technique of packet pairs is used in this case to determine the bandwidth

Each node sends a back-to-back probe packet of sizes 137 bytes and 1137 bytes to each of its neighbor every minute

The neighbor measures the time difference between the receipt of the first and the second packet and communicates it back to the sender

The sender takes the minimum 10 consecutive samples and estimates the bandwidth by dividing the size of the second probe packet by the minimum sample

N1

Sender

N2

N3

N4

P1 P2

P1 P2

P1 P2

P1 P2

Implementation of MR-LQSR

Implemented in an ad-hoc routing framework called the Mesh Connectivity Layer (MCL)

MCL is a loadable windows driver and implements a virtual network adapter within

To the rest of the system, the ad-hoc network appears as an additional network link

It internally routes the packets using the LQSR protocol

IPv4 IPv6 IPX …

MCL (with LQSR and WCETT)

Ethernet 802.11 802.16 …

Note: The above diagram has been borrowed from [1]

Implementation - Advantages

Higher layer software runs unmodified over the ad-hoc network. Hence, no modification to the network stack is required

The virtual MCL network adapter can multiplex several physical network adapters. Hence, the ad-hoc routing runs over heterogeneous link layers.

Testing

The implementation has been tested on a testbed consisting of 23 wireless nodes

The testbed is located in an office floor and the nodes are placed in cubicles, conference rooms and labs

All nodes are HP machines with latest configuration and with Microsoft Windows XP as their operating system

Each node has two 802.11 radios connected to the PC via PCD-TP-202CS PCI-to-Cardbus adapter cards and each node has a NetGear WAG 511 or NetGear WAB 501 card

Testbed…

Note: The above diagram has been borrowed from [1]

Results

The results have been classified as– Accuracy of bandwidth estimation– Baseline scenario – Single radio– Two radios– The impact of β– Two simultaneous connections

Results - Accuracy of bandwidth estimation

Two of the testbed nodes were used

The time between successive pair of packets was 2 seconds

Each bandwidth estimate was obtained by taking the minimum of 50 such pairs

The estimation is not accurate for higher rates.

Note: The above diagram has been borrowed from [1]

Results - Baseline scenario - Single radio

Out of 506 sender-receiver pairs, 100 pairs were picked at random

A 2-minute TCP transfer was carried out between the selected pair of nodes

The experiment was carried out for WCETT, ETX and for basic shortest-path routing

Since each node had a single radio, the throughput difference between the three protocols were not that significant

Note: The above diagram has been borrowed from [1]

Results – Two radio

One 802.11a radio and one 802.11g radio per node was used

The same TCP transfer was used with the parameter β set to 0.5 for WCETT

As shown in the figure, WCETT outperformed the other protocols by a huge margin

This is due to the fact that WCETT takes into consideration the channel diversity of the link too in addition to bandwidth of the link

Note: The above diagram has been borrowed from [1]

Results – One and two radios

Note: The above diagram has been borrowed from [1]

Results - The impact of β

β plays an important role in the WCETT calculation

When β is set to 0, WCETT selects the link based only on the ETT or the latency, without regard to the channel diversity

Setting the value of β to 1 makes little sense

The metric selects the paths with less channel diversity when β is low

Note: The above diagram has been borrowed from [1]

Results - Two simultaneous connections

For WCETT metric, the experiment was repeated four times with β = 0, 0.1, 0.5 and 0.9

The measured median throughput was multiplied by 2 since there were two connections. The product was called the Multiplied Median Throughput (MMT)

It must be noted that WCETT performs better than ETX for all values of β

The conclusion is that at higher loads, the throughput is maximized by having lower values of β

Note: The above diagram has been borrowed from [1]

Related work

One way to improve the capacity of wireless networks is by using improved MAC

– To exploit multiple non-interfering frequency channels An alternative way to improve the capacity is to

stripe traffic over multiple network interfaces Another approach is to use directional antennas The capacity of wireless network can also be

improved by taking advantage of full spectrum by using rapid channel switching

– This can be quiet slow with the existing hardware– Can be implemented if hardware support is achieved

Conclusion

It is shown that when nodes are equipped with multiple heterogeneous radios, it is important to select channel diverse paths in addition to taking care of latency and bandwidth for links

The results show that WCETT outperforms the existing protocols in this particular scenario where channel diversity is involved

WCETT is flexible in the sense that it allows us to tradeoff the channel diversity by setting the value for β

The implementation calls for no change in hardware or the networking software. This allows the user to seamlessly use this protocol with the existing system setup

References

[1] Richard Draves, Jitendra Padhye and Brian Zill “Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks”

[2] D. De Couto, D. Aguayo, J. Bicket, and R. Morris: "High-throughput path metric for multi-hop wireless routing", In MOBICOM, 2003.

Questions, corrections and suggestions?

Thank you