Ad Hoc and Mesh Networks: Architecture and Technology Overview - - PowerPoint PPT Presentation

1

Ad Hoc and Mesh Networks: Architecture and Technology Overview

Rutgers, The State University of New Jersey

• D. Raychaudhuri

ray@winlab.rutgers.edu www.winlab.rutgers.edu

2

Introduction: The Mesh Network Opportunity ….

3

Ad Hoc and Mesh Networks: 1st Gen Products

From Firetide

Peer-to-peer network that allows groups

• f nearby users to communicate,

exchange files, stream media, work collaboratively, …

Ad Hoc Mesh

4

Ad Hoc and Mesh Networks: Background

• Several distinct motivations for ad hoc and mesh
• Connecting ad hoc cluster of mobile users (tactical, vehicular, P2P)
• Networks involving embedded low-power devices (sensor nets)
• Access without wired infrastructure (rural, developing countries)
• Short-range radio cost-performance wide area
• 1st generation products were for specialized markets
• Tactical, specialized ad hoc applications
• Sensing applications with power constraints
• 2nd gen products are for existing telecom markets, exploiting

exceptional cost-performance of commodity radios…

• Initially rural telecom, hobbyists metro mesh today
• Now migrating to mainstream broadband access
• Is cellular next?

5

Ad Hoc and Mesh Networks: The PC Analogy

Wired High-Speed Network (Ethernet Switch or Internet) Wired High-Speed Network (Ethernet Switch or Internet) Mainframe Computer Distributed PC’s Cellular BTS Tower Networked Low-Cost Radios ~$10K/GIPS ~$0.5K/GIPS – cheap but uncoordinated CPU cycles Distributed PC solution dominates for most regimes except supercomputing

Technical issues: communication latency, overhead, parallel computation issues, execution control, unreliable networks, etc. mostly solved!

Lower cost, higher capacity, more robust?? ~$1M/Mbps (long-range) ~1K/Mbps – cheap short-range but uncoordinated basic transmission

??

Distributed PC solution dominates for most regimes except supercomputing Lower cost, higher capacity, more robust??

Technical issues: communication latency, overhead, Concurrent transmission issues, network control, unreliable channels, etc. not solved yet!

6

Ad Hoc and Mesh Networks: The PC Analogy (contd.)

The $49 Mesh Node from Meraki Networks*!

1000 node metro mesh would cost just ~$50K in capital to cover a ~10 sq-Km area…!!

*Stanford and MIT student startup

7

Ad Hoc and Mesh Networks: Product Space

Radio Range End-User Service Bit-Rate 1 m 10 m 100 m 1Km 10 Km 10 Kbps 100 Kbps 100 Mbps 1 Mbps 10 Mbps

2G Cellular 3G Cellular Access Networks (WiMax) Metro Mesh 802.11b 802.11 a,g Indoor Mesh LAN Wide-area access WLAN office/home and campus access Dense

• ffice

& home access UWB Dense AP Mesh Regime Tactical Ad hoc net

• Mesh extends 802.11x

radios to cover:

• Metro mesh (medium

range, high capacity)

• Access networks

(extended range, lower capacity)

• Indoor WLAN (higher

capacity, coverage)

• Region of use can be

even greater with new non 802.11 radios

• Cellular wide-area

equivalent

• Switched Ethernet

equivalent indoors

8

Ad Hoc and Mesh Networks: 2nd Gen Products

Dual-radio ad-hoc router (includes wired interface for AP sites) (above photo shows WINLAB’s ORBIT node) Radio Nodes ~50-100 m spacing Ad-hoc Radio links Access Point (wired) Ad-Hoc Radio Node

Office WLAN (faster, more scalable) than current 802.11 Metro Area Mesh Network (dense, high capacity, low cost)

Commercial vendors: Tropos, Motorola, Nortel, Nokia, … Commercial vendors: Firetide, Cisco, …

9

Ad Hoc and Mesh Networks: Problems with Current Technology

1st gen, and to some extent, 2nd gen solutions

suffer from several technical problems:

• Poor scalability – too many hops!
• CSMA/CA MAC implies “exposed nodes” which cannot tx in parallel
• MAC protocols never designed for multi-hop wireless to begin with!
• Topology changes rapidly – increases routing overhead
• Overall control overhead can be very high
• Routing unaware of changes in PHY speed/quality
• “Self-interference” effect for TCP flows

10

Ad Hoc and Mesh Networks: Technology Issues

Products are being released, but…

• Mobile ad hoc networks don’t really work well for tactical & vehicular
• Mesh network performance is marginal, OK for low-cost scenarios only

Major technical challenges are

• Scalability (overcoming Gupta & Kumar) – hierarchies, multi-channel
• PHY capacity improvements – collaboration, MIMO, network coding
• Topology discovery and self-organization in mobile scenarios
• Reducing control overheads in existing 802.11 MAC and MANET routing
• Mitigating MAC “exposed node” problem for parallel transmissions
• Integrated or cross-layer MAC & routing approaches without the

performance problems of conventional layered protocols

• Introducing service features such as QoS, multicast, …
• Network Security!

WINLAB research covers several of the above topics...

11

Key Technologies

12

Key Technologies: Hierarchical Architecture

• Hierarchical structure essential for scalability
• Classical “Gupta & Kumar” result shows mesh throughput per node does down as sqrt(n)
• System can scale with multiple frequencies and proper ratio of MN, FN and AP
• E.g, if MN~100, ~10 FN’s & ~3 AP’s needed (…note significant reduction in # wired nodes)

Wired Internet Infrastructure

Gateway node Potential bottleneck

“Flat” mesh network with ad-hoc routing: does not scale! Wired Internet Infrastructure

Multi-tiered Interfaces to wired network

Hierarchical architecture with multi-radio forwarding nodes and AP’s

Ad-hoc associations Ad-hoc associations

Throughput per node scales ~ 1/sqrt(n) Throughput per node scales with right ratio of FN’s, AP’s

Grid Portals/ Access Points

Multi-radio Forwarding Node

Sample experimental result on ORBIT showing linear scaling & ~2.5X capacity (for a mesh network with ~20 MN, 4FN, 2AP)

13

Transmit Power Required: 1mW Assoc Req Transmit Power Required: 10mW

FN

AP

• Scan all channels, record neighbors
• Decide neighbor based on objective
• Associate with neighbor

Key Technologies: Discovery and Self-Organization

• Only a subset of available links made available to routing – achieves

balance between routing overhead and route availability

• Dynamic topology formation based on different such as max

throughput, min delay or power

Send beacons

FN

AssocReq AssocReq

Logical topology

FN

Interface Two Send beacons Accept Associations Forward client Data Interface One Scan all channels Find minimum delay links to AP Associate with AP

Wired Internet Infrastructure

PHY MAC

DISCOVERY

ROUTING

Sample Result showing significant reduction in routing

• verhead

14

200 400 600 800 1000 1200 100 200 300 400 500 600 700 800 900 1000 O ffered Load (kbps) System throughput (kbps) S ystem throughput (S cenario I) M H M etric PA R M A

500 1000 0.5 0.6 0.7 0.8 0.9 1 Offered Load (kbps) Packet Delivery Ratio Packet Delivery Ratio (Scenario I) MH Metric PARMA 500 1000 100 200 300 400 500 600 700 Offered Load (kbps) End-to-End Delay (ms) End-to-end Delay (Scenario I) MH Metric PARMA

Improved performance with PARMA compared to MH metric. PARMA has the same behavior as MTM when no congestion.

Routing Metric = Σ pkt size/link speed + MAC congestion

Key Technologies: Cross Layer Routing

15

Key Technologies: Multi-Channel Mesh

Multi-channel mesh (>>1 radio per node) can improve performance

significantly by supporting concurrent transmissions & reducing/eliminating 802.11 MAC overheads

• Many 2nd gen mesh products use 5-6 radios per node

Algorithms for optimizing throughput given constraints on # radios, # channels Possible to use 802.11a hardware and avoid MAC effects entirely

f1 f2 f3

16

Key Technologies: Clean-Slate PHY/MAC for Mesh

• Wideband, agile short-range OFDM radio design optimized for speed
• For example, ~2 x 20 Mhz bandwidth, max bit-rate ~250 Mbps
• Additional low-bit rate PHY for control, flexible TDMA based MAC
• Radios can switch channels and bit-rates on a slot-by-slot basis (~ms) –

allows for unconstrained FD/TDMA allocations

2x20 Mhz Agile RF Front End

D/A

OFDM Baseband MAC control Interface ~25-200 Mbps Service data Programmable radio board at WINLAB

2.4 Ghz RF 1 Mbps 802.11b PHY

Control Plane data

PHY Module #1 PHY Module #2 Control PHY Computing Module

Grid Node Platform Note: 2 x 200 Mhz agile radios with TD capability should be sufficient for ~50 mbps duplex per node Assuming ~3-4 hops to a wired AP

17

Key Technologies: Global Control Plane

• Important architectural idea – clean separation of control & data planes
• Reduces control overhead and enables contention-free global

MAC/routing algorithms

• Can use a single low rate channel (e.g. 1 mbps 802.11b) for control

http://www.uninett.no/wlan/throughput.html

Example of WLAN throughput breakdown

Control penalty In current WLANs (increases with bit-rate) Low Bit Rate PHY Bootstrapping/ Discovery Link stats Flow stats

Control Plane Data Plane

Fast/Agile PHY Lower MAC Integrated MAC/Routing

Frequency assignments, TDMA schedule and route selection

Distributed or Global Control Integrated MAC/Routing Algorithm Broadcast MAC & Routing

Control Plane Data Plane

Ethernet and 802.11 drivers

18

Key Technologies: Integrated Routing and MAC

• Global allocation of routes and

MAC time slots at the same time to completely eliminate contention

• Allocation algorithm works on

both frequency (FD) and time (TD)

• Algorithm checks for compatible

time slot and freq at each receiver

• Allows for more parallel

transmissions (fewer “exposed nodes”) and eliminates packet contention

• Significant performance

improvement over conventional layered 802.11 + AODV etc.

• Requires GCP-type capability for

distribution of control

Comparison of Individual and Aggregate Throughput

200000 400000 600000 800000 1000000 1200000 1400000 flow 1 flow 2 flow 3 flow 4 flow 5 Total Global Scheduling 802.11 Aloha Slot Aloha

19

Key Technologies: Cognitive Radio

A

A

B

B

D

C D E F

Cognitive radios provide PHY, MAC flexibility needed to implement

cooperative multi-channel ad hoc networks with better performance

• Cognitive radio can achieve multi-channel + flexible MAC performance
• PHY can be optimized between neighboring nodes
• High spectrum efficiency possible via dynamic spectrum algorithms
• Networks may utilize a control channel similar to GCP, etc.

Bootstrapped PHY & control link End-to-end routed path From A to F PHY A PHY B PHY C Control (e.g. CSCC) Multi-mode radio PHY Ad-Hoc Discovery & Routing Capability

Adaptive Wireless Network Node (…functionality can be quite challenging!)

~250 Mbps Control Plane

20

A B C D E R1 R2 R3 R4 R5 rejected rejected flow 1 (t1) flow 2 (t2) flow 3 (t3) request Response bw resv ::: W Hops=0 ::: Forwarding Set a timer Check bandwidth availability W N hops B

cont up consume

× + + = ) 1 2 , 1 min( drop ::: W Hops=0 ::: Resv Bw Set a timer forwarding Check bandwidth availability

up consume cont consume

B W N hops B + × = ) 2 , min( drop

• Establishes QoS routes with reserved bandwidth on a per-flow basis
• Monitors interference from adjacent nodes
• Performs admission control to maintain network utilization below the

congestion point

Key Technologies: QoS-Aware Routing

21

• Location is a more

natural addressing mechanism

• Location becomes more

important than a network address

• Opportunistic

message forwarding within geographic perimeter

• Retransmissions

from different vehicles

• Delay-tolerant

networking

Desired message delivery zone (Idealized) Broadcast range Irrelevant vehicles in radio range for few seconds Passing vehicle, in radio range for tens of seconds Following vehicle, in radio range for minutes

Key Technologies: Geocasting in Ad Hoc Vehicular Networks

22

Experimental Results

23

Access Point

US Robotics 2450 AP AMD Elan SC400 processor 1 MB Flash, 4 MB RAM Prism-2 based PCMCIA card

Forwarding node

Compulab 586 CORE AMD Elan SC520 CPU 2 MB NOR flash + 64 MB NAND Flash on board Dual PCMCIA slots

Sensors

Intrinsyc Cerfcube Intel PXA 250 (XScale processor) CF-based wireless support HARDWARE PLATFORM SOFTWARE

802.11b ad-hoc mode

Experimental Results: WINLAB Prototype circa 2002

24

Experimental Results: ORBIT Radio Grid

• ORBIT: 400 nodes in 20m x 20m– two 802.11 radios each (atheros and intel-based)
• Intended for ad hoc and mesh network studies

Antennas Mini ITX-based SSF PC

25

Urban 300 meters 500 meters Suburban 20 meters ORBIT Testbed

20 meters

Hallway Office

30 meters

Experimental Results: Radio Mapping

26

Experimental Results : ORBIT Proof-of-Concept for Metro Mesh Scenario

Flat Hierarchical

System Parameters: 0.9 sq. km, 20 mobiles/sensors, 4 FNs, 2 APs 802.11a with multiple frequencies

15 20 25 30 35 40 45 50 55 60 65 10 15 20 25 30 35 40 45 50 System offered load (Mbps) System Throughput (Mbps) Total System Throughput for flat and hierarchical topologies Flat Hierarchical

Flat Hierarchical

• “SOHAN” system evaluated for realistic

deployment scenario with ~25 nodes

• Results show that system scales well

and significantly outperforms flat ad-hoc routing (AODV)

AP FN MN

Mapping on to ORBIT Radio grid emulator

27

Mapping of the actual path onto ORBI T.

Emulated path. Emulated path. Actual path. Actual path.

A B C D E F H G

Experimental Results: Mobility Emulation for MANET Studies

Goal: Emulate mobility for MAC and higher layers for larger number of nodes

28

Objectives

Demonstrate a vehicular 802.11a experiment. Actors : Sender, Receiver Details : Receiver is stationary. Sender moves around the parking lot. Sender transmits ICMP packets addressed to the receiver. Both nodes use 802.11a, channel 36. Receiver logs per-packet RSSI using Libmac.

Results (Snapshot of RSSI) Experimental setup

Experimental Results: ORBIT Vehicular Setup

29

Experimental Results: Ongoing Work

IRMA – Integrated Routing and MAC (..requires

software MAC capability under development)

Includes a global control plane (GCP) Centralized and distributed control algorithms to be compared

Cross-layer routing with DCMA cut-through

switching MAC

Switched multi-radio mesh scenario Based on DCMA (Acharya) MAC, distributed cross-layer routing

Vehicular ad hoc network scenarios

Dense MAC experiments Geocasting protocol evaluation