Calculating a Next-hop Metricin Multiradio Networks

Ian Chakeres and Guntur Ravindra

Motorola Labs

Bangalore, India{ian.chakeres,ravindra.guntur}@motorola.com

Abstract-With the proliferation of multiple radio de­vices, simply choosing to send all traffic over the onenext-hop radio link with the lowest metric is foolish.Unfortunately, this behavior is what exists today. In thispaper, we describe how to combine multiple radio links'metrics to a particular next-hop device. The combinednext-hop metric allows a multiple radio device to choosethe best next-hop device for traffic considering all theradio opportunities available. By considering all radioopportunities, performance may be improved significantly.

I. INTRODUCTION

Today's wireless devices have multiple radios.For example, many mobile phones today often havethree radios. As shown in Figure 1, a mobile phoneoften has a cellular radio, a IEEE 802.11 radio [1],and a bluetooth™ [2] radio. Unfortunately, mostrouting protocols consider each link, and the next­hop device on the other end of the link, indepen­dently. Therefore, only a single radio link will beutilized for all traffic.

The fact that each link, and its metric, is con­sidered independently is foolish, since there areoften multiple parallel radio links connecting twodevices. For example, two mobile phones may beable to communicate over both an IEEE 802.11radio and a bluetooth™ radio simultaneously. Inthis scenario, both radio links could be utilized toimprove the communication performance betweenthe two devices.

The first step to enabling two devices to utilizemultiple parallel radio links is to develop a com-

cellular

bluetooth 1M

Man

Figure 1. Mobile phones exemplify the prevalence of

multiple radio devices.

bined next-hop metric that can be compared withother devices' next-hop metric. The routing protocolcan then use the combined metrics to choose thebest next-hop among several opportunities; in thesame way routing protocols currently use link orpath metrics.

The main contribution of this paper is the deriva­tion of this combined multiple radio next-hop metricthat allows a routing protocol to determine the bestnext-hop considering all radio opportunities.

The rest of this paper is organized as follows.Section II describes traditional routing protocolsnext-hop decision, parallel links, and how parallellinks are handled. In Section III we formulate thenext-hop metric challenge and present our solutionto allow the routing protocol to make an intelligentnext-hop decision. We also discuss how routing pro­tocols can utilize the combined next-hop metric andother implementation details in Section IV. FinallySection V summarizes the main contributions andelaborates on our future follow-on research.

Figure 2. Traditional routing protocol next-hop deci­

sion example

II. BACKGROUND

A. Routing Next-hop Decision

Routing decisions are normally based on linkor path metrics. For example, OSPF [3] considerseach link's metric while calculating the shortestpath between two OSPF routers. RIP [4] routingdecisions are based upon the total distance betweenRIP routers. In both, scenarios the route with thelowest (cost) metric is selected.

To help describe how traditional routing happensin a network where each link has a metric, weprovide a simple example. Figure 2 shows a smallnetwork where each link is labeled with its linkmetric. In this example, node X needs to determinewhich route will be used to reach node D. Thereare two paths between node X and node D; onevia node Y and another via node Z. The route cost(sum of link metrics) of the route between node Xand node D via node Y is 6 + 1 == 7. The costof the route via node Z is 4 + 1 == 5. Therefore,the routing protocol will choose to use the lowercost route via node Z for communication betweennode X and node D.

B. Parallel Links

As described in the introduction, today's devicesoften have multiple radios and these radios formparallel links with other nearby devices. Figure 3shows two devices (node X and node Y) withmultiple parallel radio links between them. Each ofthe radio links has an associated metric mn .

2

Figure 3. Two devices with n parallel links between

them. Each link is labeled with its link metric, m n •

c. Traditional Handling ofParallel Links

One problem with traditional routing is that byconsidering each link independently only one ofmultiple parallel links will be utilized. For example,Figure 4 shows two devices with multiple parallellinks. The top link has a metric of 1 and thelower links have metrics 2 and 3. In this example,traditional routing will always choose the lowestcost link (the upper solid-line link) and the otherlinks (the lower dotted-line links) will be ignored.These other links with higher costs will be left un­utilized.

Another problem with routing protocols consid­ering each link individually is shown in Figure 5.In this figure, node X is again selecting a routeto reach node D. Notice that in this figure anotherlink has been added between node X and node Y.Since the route via node Z still contains the smallestroute metric, node Z will be selected. Although,using both links between node X and node Ymight actually be more efficient and allow higherperformance.

These examples highlight the reason for creatinga combined next-hop metric. If a combined next-hop

12................................

...............~ .Figure 4. Traditionally only the best link (solid-line) is

utilized; other links (dotted-lines) are left un-utilized.

3

(3)

(1)

(2)

Before we can develop the combining function,we must first define the meaning of a link's metric.In many wireless networks the link metric is pro­portional to the throughput of the link. Therefore,we define the link metric, m n , to be the amountof time it takes to transmit one unit of data onlink n between node X and node Y. This linkmetric is used in many wireless networks. For ex­ample, this link metric is used in IEEE 802.11 s [5].This metric is also very similar to ETX [6] andWCETT [7].

This link metric is good because it is able toabstract and incorporate many different wirelessradio factors. For example, this link metric is ableto abstract the physical transmission data rate, theframe error rate, the number of retransmissions, andthe medium access time.

Now that the meaning of the link metric hasbeen defined, we can derive a function to combinemultiple link metrics properly.

i. We start with the amount of time it takes totransmit one unit of data on link n between node Xand node Y

ii. Inverting Equation 2, we get the amount ofdata that can be transmitted in one unit of time onlink n between node X and node Y

1

III. CALCULATING A COMBINED NEXT-HOPMETRIC

The goal of creating a combined next-hop metricis to enable routing to make intelligent next-hopdecisions taking advantage of parallel link oppor­tunities.

Figure 7 shows our system, two nodes withmultiple parallel links. These parallel links are thencombined using a function to create a combinednext-hop metric, shown in Equation 1.

m

Figure 5. This figure shows what happens when tradi­

tional routing treats each link independently. In this

example, although there are parallel links between

node X and node Y they are each considered in iso­

lation. Therefore, the lowest metric route via node Z is

selected.

Figure 6. Combined next-hop metric

metric (shown in Figure 6, m == f(ml' m2, ... , m n ))

were available, then the routing protocol coulddetermine the best next-hop device while takingadvantage of parallel links.

The task of creating a combined next-hop metricis not trivial. To illustrate the difficulty we mustexamine Figure 5 again. In this example, the routingdecision boils down to choosing the next-hop nodewith the lowest metric. Given that there are two par­allel links between node X and node Y, calculatinga combined metric that incorporates the link metricsof both links is not straight-forward. For example,right now we cannot say that the combined metricfor 6 & 7 results in a lower metric than 3. In the nextsection, we describe our proposal for calculating acombined next-hop metric.

4

Figure 8. Combined next-hop example

node D. Notice that by using the combined next­hop metric the route to node D via node Y has alower cost than the route via node Z.

IV. DISCUSSION AND FUTURE WORK

In addition to using the combined next-hop metricfor comparing and deciding upon which route totake, the forwarding of data packets must be splitacross the multiple parallel links to result in a per­formance improvement. This procedure has severalimplications, some which are discussed in [8].

Strategies for splitting traffic among parallel linksmay introduce overhead and complexity. For linkswith high cost metrics the overhead of each packettransmission could actually result in a performancedecrease. Therefore, in the future we may need tointroduce an additional bias against utilizing parallellinks that do not improve performance.

Similarly, splitting traffic among parallel linksmay result in other undesirable behavior. For ex­ample, out-of-order packet delivery and high jitter.Furthermore, these effects may become more pro­nounced over multiple hops.

Further, to achieve a performance improvementour approach requires a technique wherein traffic issplit across links proportionally to the link capacity.With jitter in the network, the extent of fairness willalso need to be evaluated.

Many techniques exist to split traffic in thismanner among multiple parallel links. The splittingtechnique could involve operations that are typicallyused in a fair weighted round-robin scheduler. Alter­natively, a simple method utilizing a single queue

(4)

(5)

(6)

m

1 116m=" 1 = 1+1 =I=2=3

~mn 66 6

Using the combined next-hop metric, node Xchoose the lowest cost route via node Y to reach

iv. Inverting Equation 4, we find the combinedamount of time it takes to transmit one unit of datausing all links between node X and node Y

1

E~n

A. Combined Next-hop Metric Example

To illustrate how the combined next-hop metric iscalculated, we provide an example. Figure 8 showsa network with parallel links between node X andnode Y. Each of these links has a link metric of 6.

First we calculate the combined next-hop metric,for the links between node X and node Y usingEquation 5. The outcome is a combined next-hopmetric with value 3; the calculation's intermediatestates are shown in Equation 6.

Figure 7. System

iii. Summing Equation 3 across all links, we havethe combined amount ofdata that can be transmittedin one unit of time using all links between node Xand node Y

5

radio links' metrics into a single value that canbe compared with other link and next-hop metrics.By using the combined next-hop metric, deviceswith parallel links can attain higher network per­formance.

In the future we plan to gather experimentalresults using the combined next-hop metric in a mul­tiple radio wireless mesh network. We hope theseresults will help us quantify the cost of splittingpackets across multiple links and whether out-of­order delivery occurs often.

REFERENCES

[1] IEEE Computer Society, "IEEE 802.11 Standard, IEEE Standard

For Information Technology," 1999.

Bluetooth™SIG, ''Bluetooth™ ," http://www.bluetooth.com.

J. Moy, "OSPF Version 2," RFC 1058, April 1998.

C. Hedrick, "Routing Information Protocol," June 1988.

IEEE Computer Society, "IEEE 802.11s Standard (draft Dl.10),

IEEE Standard For Information Technology," 2008.

[6] D. S. J. De Couto, D. Aguayo, and J. B. an.J.tobert Morris, "A

High-Throughput Path Metric for Multi-Hop Wireless Routing"

in Proceedings of the 9th ACM International Conference onMobile Computing and Networking (MobiCom '03), San Diego,

CA, September 2003.

[7] R. Draves, 1. Padhye, and B. Zill, "Routing in Multi-radio

Multi-hop Wireless Mesh Networks," in Proceedings of the10th Annual International Conference on Mobile Computing andNetworking (MobiCom) , September 2004.

[8] D. Thaler and C. Hops, "Multipath Issues in Unicast and

Multicast Next-Hop Selection," RFC 2991, November 2000.

v. CONCLUSIONS

In this paper we derived a combined next-hopmetric (shown in Equation 5) that can be used byrouting protocols to make next-hop routing deci­sions. This equation aggregates multiple parallel

for each next-hop node that dequeues packets fortransmission whenever a radio opportunity arisescould be used.

In regard to energy, utilizing multiple radios si­multaneously can increase the rate of energy con­sumption; in contrast to schemes which utilize onlya single radio or a small subset of radios.

In regard to routing, combining multiple linksinto a single next-hop has the advantage of reducingthe number of routing possibilities. This reductionmay substantially reduce routing protocol computa­tion and lead to faster convergence.

In the future, we may also explore incorporatingheterogeneity into the combined next-hop metric.Heterogeneity comes in many forms: link-type, [2]

channel/frequency, traffic-type, etc. For example, if [3]

policy dictated that particular traffic should avoid [4]

traversing certain reserved radio frequencies. In this [5]

case, the combined next-hop metric might need tobe biased to give a higher cost metric for routingand carrying that traffic.