A Radio Multiplexing Architecture for High Throughput Point to Multipoint Wireless Networks Ramakrishna Gummadi Rabin Patra, Sergiu Nedevschi, Sonesh Surana,

A Radio Multiplexing Architecture for High

Throughput Point to Multipoint Wireless Networks

Ramakrishna Gummadi

Rabin Patra,Sergiu Nedevschi, Sonesh Surana, Eric BrewerUC Berkeley

WiNS-DR 2008

MIT CSAIL

A radio multiplexing architecture

Architecture (noun): the manner in which the components of a computer or computer system are organized and integrated

- For what particular wireless configuration?- Why do we care about a radio architecture?

19 September 2008 Networked Systems for Developing Regions (NSDR)

2

18 August 2008 Networked Systems for Developing Regions (NSDR)

3

Rural network connectivity What is the need?

Divide: Rural vs urban, Intranet vs Internet

Applications: Health, Education, Information access

Requirements for rural networks Low cost per user Good performance (throughput) Grassroots deployment and management Scalable expansion


4

Typical rural scenario

Com3Com3Com3

Village

City

Optical Fibre uplink

City

Village


5

Typical rural scenario

Com3Com3Com3

Village

City


City

Village


6

Point-Multipoint (PMP) networks

Com3Com3Com3

Village

City


City

Village


7

PMP features Base station:

Multiple sector or steerable antennas Multiple radios

Client: Single radio Directional antenna

Distances: Up to 20km

Traffic: Demands are time-varying and bursty

Why a radio architecture for PMP? Point-point (P-P) links needs high

throughput Know how to do this well for P-P (e.g., 2P,

WiLDNet) But cannot extend to PMP directly

System as a whole susceptible to interference Maintaining links tedious and error-prone Incremental scalability hard Inflexible to bursty traffic Most importantly, high total cost of ownership


8

High cost?


9

Cost: $70,000

Cost: $3,000

In relative GDP terms, costs can be comparable!

Towers are the hidden costRequirements Low cost per user Good performance

(throughput) Grassroots

deployment and management

Scalable expansion

Status quo Large initial costs Interference lowers

throughput Expensive and tedious to

realign or troubleshoot Adding capacity and links

impossible once “maxed out”


10

Goal

Design and evaluate high-throughput yet low-cost radio multiplexing architectures for PMP n/w


11

Where is the architecture?


12

.

.

.

Architecturegoes here

Cheap$$,

lower interference with larger sector separation


13

Talk outline Why multiplexing architecture?

Architectural principles and implications

Evaluation


14

5

6

21

4

12

3

11

10

8

9

7

Single sector scenario Clients: c1 ,c2 …cn

Single base-station


15

Multiple-sector scenario (today)

Clients: c1 ,c2 …cn

Radios: R1 ,R2 …Rm

Sector antennas5

6

21

4

12

3

11

10

8

9

7

Ch: 1

Ch: 1

Ch: 1

But interference can kill


16

A

B

C

1

2

Simultaneousreceive

α A

B

C

1

2

Simultaneoussend

α

α should be large enough!

Principle 1: Separate channels for more degrees of freedom


17

Clients: c1 ,c2 …cn

Radios: R1 ,R2 …Rm

Sector antennas Each sector on

different channel Both

directional and frequency separation gains

5

6

21

4

12

3

11

10

8

9

7

Ch: 1

Ch: 2

Ch: 3

Principle 2: Exploit spatial reuse

Multiple channels per sector antenna

Channels as widely separated as possible

Spatial diversity and multiplexing gains


18

5

6

21

4

12

3

11

10

8

9

7

Principle 3: Use cheap h/w to increase capacity


19

Wireless cards cheap

Commodity splitters and combiners cheap

Linear capacity increase possible

But ensure sufficient RF isolation!

RF isolation Isolation from

commodity splitters may not be enough

TDMA MAC solves this problem nicely


20

Tx

Rx

Principle 4: Allocate radios dynamically for bursty traffic

Client traffic is bursty Static radio assignment

sub-optimal A multiplexing

controller after splitter switches radios to clients dynamically

2/4-port muxers affordable; higher port counts lossy and costly


21

Key architectural benefits

Number of sectors: S Number of orthogonal

channels: C

Total #antennas: S*C

Peak #clients per sector: C

After Still S*C cards, but:

Total #antennas: S Towers can be smaller

Peak #clients per sector: S*C Greater spectral efficiency So, more throughput per

client, or more clients


22

Before

Additional benefits Low cost per user

Fewer antennas with more channels and radios Grassroots deployment and management

Shorter towers means easier alignment If radio or link fails, switch to under-used or

spare Scalable expansion

New clients added by allocating radios permanently


23


24

Talk outline Why multiplexing architecture?

Architectural principles and implications

Evaluation

Evaluation 3 clients, 3 PMP links Radios: 25 dBm

max. 3-way muxer, 20 dB

isolation 20 dB attenuators Metrics:

Simultaneous Tx/Rx, Tx+Rx throughput

Effect of channel separation, isolation


25

Aggregate UDP throughput


26

#Radios

Channels

All radios

Tx

All radios

Rx

MixTx/Rx

2 1,6 14.51 14.89 14.77

2 6,11 14.74 13.98 14.16

2 1,11 13.90 13.79 13.70

3 1,6,11 21.14 20.34

Muxing works as expected for 3 radios, even for Rx/Tx

Throughput vs. isolation


27

At sufficient channel separation and isolation, aggregate CSMA throughput unaffected.

Need for TDMA otherwise.

Conclusions Radio multiplexing can reduce large-towers

Maintains throughput and simplifies management Commodity splitter and combiners can be

used Need to think about RF isolation carefully Cost<->complexity trade-off can be hard

At PHY layer, complementary to WiLDNet Future work

Look into actual deployments TDMA MAC must synchronize Tx/Rx across radios


28


29

Thank You!

A Radio Multiplexing Architecture for High Throughput Point to Multipoint Wireless Networks Ramakrishna Gummadi Rabin Patra, Sergiu Nedevschi, Sonesh Surana,

Documents

regions nsdr18

regions nsdrtowers

regions nsdrwhy

regions nsdrgoaldesign

regions nsdrwhere

burstynetworked systems

regions nsdrhigh cost

mit csailnetworked systems