Improving Throughput in Multihop

Wireless NetworksZongpeng Li and Baochun Li, Senior Member, IEEE

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY 2006

Advisor: Yeong-Sung, Lin Presented by Yen-Yi, Hsu OPLAB, IM, NTU.

OPLab, NTUIM

Author

Zongpeng Li

received the B.E. degree in computer science and technology from Tsinghua University,Beijing, China,in 1999the M.S. degree in computer science and the Ph.D. degree in electrical and computer engineering from the University of Toronto, Toronto, ON, Canada, in 2001 and 2005

He is currently an Assistant Professor at the Department of Computer Science, University of Calgary, Calgary, AB, Canada. His research interests are in data networks and distributed algorithms.

2

OPLab, NTUIM

Author

Baochun Li (M’00–SM’05)

B.Eng. Degree in computer science and technology from the Department of Computer Science and Technology, Tsinghua University, Beijing, China, in 1995 The M.S. and Ph.D. degrees in computer science from the Department of Computer Science, University of Illinois at Urbana-Champaign, in 1997 and 2000an Associate Professor in Department of Electrical and Computer Engineering, University of Toronto, Toronto, ON, Canada

held the Nortel Networks Junior Chair in Network Architecture and Services since October 2003 and the Bell University Laboratories Endowed Chair in computer engineering since August 2005. interests include application-level QoS provisioning and wireless and overlay networks.a member of the ACMHe was the recipient of the IEEE Communications Society Leonard G. Abraham Award in the Field of Communications Systems, in 2000.

3

OPLab, NTUIM

Agenda

Introduction

Preliminaries

Data Dissemination

Data aggregation

Conclusion

4

OPLab, NTUIM

Introduction

5

OPLab, NTUIM

Introduction(1/2)

Broadcast nature of wireless ad hoc networks lead to interferences and spatial contention

Flows also compete for shared channel bandwidth if they are within the transmission ranges of each other (contention in the “spatial domain”)

Broadcast may be of assistance in the multicast scenario

Maximum achievable throughput depends on1.Arranging network topology between source and

destination2.Activating per-node algorithms such as network

coding

6

OPLab, NTUIM

Introduction(2/2)

We discuss solutions to the problem of increasing end-to-end throughput in three cases that cover all scenarios of end-to-end communications Unicast data dissemination Multicast data dissemination Data aggregation

7

OPLab, NTUIM

Preliminaries

8

OPLab, NTUIM

Preliminaries

Two single-hop subflows of a multihop flow interfere with each other if and only if either the source or the destination of both flows are within the single-hop transmission range.

focus on multihop flows that traverse more than two hops, thus consisting of more than two subflows, because these multihop flows exhibit spatial contention even among its own subflows

9

OPLab, NTUIM

Preliminaries

One-hop away ： Nodes are not within transmission range of each other

10

OPLab, NTUIM

Preliminaries

C ： single-hop wireless channel capacityr ： the achievable throughputT ： total transmit timeS ： the set of sourcesti ： the amount of time where source i is

scheduled to transmit during the scheduling period T

r=C ∑‧ i ∈ S ti/T

11

OPLab, NTUIM

Preliminaries

Maximum contention clique(mcc) A set of links such that any two links within the set interfere

with each other with maximum size

6

12

6

OPLab, NTUIM

Preliminaries

Theorem ： r ≤ C ∑i‧ ∈ S ti/|mcc|,equality holds if transmission topology is a forest

we have r = (C · 1 + C · 1)/6 = C/3, and the achievable throughput for each session is r/2 = C/6

13

6

OPLab, NTUIM

Data dissemination

14

OPLab, NTUIM

Data Dissemination

A. Unicast ： in wireline network, the throughput of the unicast session

is able to C as well in wireline ad hoc networks, the achievable session

throughput r is only C/3, because the mcc has size 3

Due to the intraroute spatial contention, only one out of every three links can be transmitting as a given time.

15

Data Dissemination

One route is not sufficient to effectively utilize the available channel capacity at the source

Using multipath routing(the broadcasting nature) to break the bound

To achieve load balancing and fault tolerance, the set of routes being chosen are required to be disjoint or partially disjoint

However, intense contention may still exist To reduce interroute interference and achieve a

higher session throughput, the routes need to be one-hop away

OPLab, NTUIM 16

Data Dissemination

The total number of hops on both routes is a multiple of 3, all subflows can be scheduled without interference in three equal-length phases r = C ∑ i S ∈ ti/T = 2C/3

The total number of hops of subflows is not a multiple of 3 it takes four phases to schedule all of them r =C/2

OPLab, NTUIM 17

OPLab, NTUIM

Data Dissemination

Further increase the number of routes, achievable throughput can be increased when routes is not beyond 5

Because a wireless node can have at most five one-hop away neighbors

routes phases Throughput

3 4 、 5 3C/4 、 3C/5

4 5 、 6 4C/5 、 4C/6

18

Data Dissemination

In cases where a source has data to transmit to multiple unicast destinations, one-hop away multipath routing may also be applied to increase the achievable throughput

The topology is a tree|mcc|=k+1r = C ∑ i S ti/|mcc| = kC/(k + 1), for k ≤ 5∈

OPLab, NTUIM 19

OPLab, NTUIM

Data Dissemination

B. Multicastthe throughput of a single multicast session is also bounded by C/3, the throughput may be increased by activating one-hop away

20

OPLab, NTUIM

Data Dissemination

Branching point branching late v.s. Branching early

Branching early leads to waste of bandwidth rather than an improvement of throughput

Both achieves C/3 ， but (a) consumes half of the bandwidth

21

OPLab, NTUIM

Data Dissemination

Use multiple one-hop away routes to “strengthen” longer routes before late branching, the throughput may be increased

r increased to 2C/5

22

OPLab, NTUIM

Data Dissemination

C. Network Coding Usually consumes less bandwidth

Achieves a throughput of 2C

R1 receive f1 and (f1⊕ f2),get f2 by f2=f1 ⊕ (f1⊕ f2),similarly, R2 get f1 by f1=f2 ⊕ (f1⊕ f2)

23

OPLab, NTUIM

Data Dissemination

not as advantageous for smalland dense ad hoc networks

• Complicated cyclic topology

• Nonidentical data flows

The throughput is boundedby C/3

Multicast tree withoutcoding using routesS-A-R1 、 S-B-R2 is able to achieve C/2

• Coding ： C/3• Without coding ： C/2

24

Data Dissemination

The advantage of coding to increase throughput can outweigh the disadvantage introduced by spatial contention if

• Transmission network is large and sparse(|mcc| ≥4)

• Spatially nearby multicast sessions exist concurrently

OPLab, NTUIM

• Coding ： C/2• Without coding ： C/3

Each link is replaced by a multihop route

25

OPLab, NTUIM

Data Dissemination

Networking coding reduces contention by broadcast nature and by reducing the amount of data transmitted at the bottleneck link

26

OPLab, NTUIM

Data aggregation

27

OPLab, NTUIM

Data aggregation

Data corresponding observed by sensors are routed toward “data sink”

Aggregation ratio ： α α=0.5→perfect aggregation α=1→zero aggregation

Source-side amount ≥ destination –side amount

28

Data aggregation

Early aggregation v.s. Late aggregation Early aggregation reduce the overall amount of data and the total

amount of energy consumption Late aggregation is more robust Early aggregation may introduce a higher latency

OPLab, NTUIM

r = 2C/(2 +4α) = C/(1 + 2α)α [0.5, 1], r [C/3, C/2]∈ ∈

r=2C/3Similar to the 1-to-2 independent unicast case

29

Data aggregation

As the number of source flows (n) increases, the achievable throughput of late aggregation soon increases to 5C/6 where it stops

But the achievable throughput of early aggregation keeps increasing, and depending on α, it may soon become higher than C

OPLab, NTUIM 30

α [0.5, 1]∈

OPLab, NTUIM

Data aggregation

31

OPLab, NTUIM

Data aggregation

for very small n, using one-hop away routes to transmit data directly to the sink without aggregation is the best choice regardless of α;

otherwise, source flows need to aggregate to a smaller number before reaching the sink, except for cases where the α near 1

32

OPLab, NTUIM

Data aggregation

It utilize the receiving capacity at the sink quite well because: Allow source flows to aggregate into more

condensed flows before being received by the sink Allows the sink to receive data from different final

flows in an interleaving way to reduce its idling time

The number of final flows too small: single final flow forms a bottleneck too large: intense contention around the sink

33

OPLab, NTUIM

Conclusion

34

OPLab, NTUIM

Using strategies that include 1.multiple end-to-end paths 2.per-node algorithms such as coding 3rearranging transmission network topologies

It’s feasible and practical to increase data throughputAdopt the best possible strategy based on the insights in

this paper may help to alleviate such problems

35

www.themegallery.com