Transcript The WHYNET Testbed (Rajive Bagrodia)

WHYNET: Scalable Testbed for Next-Generation
Mobile Wireless Networking Technologies
R. Bagrodia, E. Belding-Royer, B. Daneshrad, M. Fitz, M. Gerla, S. Krishnamurthy,
U. Mitra, P. Mohapatra, M. Molle, R. Rao, C. Shen, M. Srivastava, M. Takai
Rajive Bagrodia
UCLA Computer Science Department
WMPG Workshop
August , 2005
Future Wireless Networks: Trends & Challenges
• Growing demand for mobile access, but wireless capacity limited
– Departure from traditional layering approach to cross-layer design
approaches (e.g., information sharing, joint design)
– Exploiting wireless link flexibility by dynamic selection of transmission
parameters (e.g., power, channel, modulation)
• Convergence and interoperability of heterogeneous systems with
diverse radio technologies (e.g., 3G, 4G, WLAN, UWB)
– Multi-interface/multi-mode/programmable radio devices
• Self-organizing wireless networks to extend coverage, reduce
costs, potentially also improve performance (e.g., mesh networks;
ad-hoc access networks)
• Embedded networked sensing and its applications
• Security issues
WMPG Workshop
August , 2005
Implication for Networked Systems
•
•
•
•
•
The next generation wireless communication technology being
developed for this purpose will be adaptive (software-defined radios,
smart antennas, programmable networks, …)
There is substantial ‘cross-layer interaction’ among the technology
solutions at multiple layers of the protocol stack (e.g., medium access,
routing, and transport) to provision dynamic Quality of Service among
the voice, video, and data traffics that must be carried by such
networks
There is limited experience, in the commercial or military arena, with
large scale deployments and use of such on-the-move communication
technology
Static analysis and planning may not be adequate to achieve the
dynamically varying Quality of Service requirements for the diverse
applications
Real-time network simulations/emulations can play a critical role in
assessing the dynamic impact of net-centricity in the design and
operation of such networks
WMPG Workshop
August , 2005
WHYNET: Thesis
•
•
•
Cross-layer interactions between radios, antenna, protocols and
applications is key to exploiting next generation wireless technologies.
A hybrid testbed that exploit advantages of physical and simulation
testbeds to study such intercations and their impact on application-level
performance
Testbed must be accurate, efficient and scalable:
– Accuracy: permit phy, radio, MAC, & networking attributes to be reflected in
a common setting
– Efficiency: real-time evaluation of models with tens of nodes.
– Scalability: evaluation of wireless networks with thousands of devices
•
•
•
Demonstrate the unique contributions of WHYNET for cross-layer
optimization studies
Generate a repository of wireless networking scenarios,
measurements, models and implementations
Make testbed accessible to the research community via the web
WMPG Workshop
August , 2005
Potential Impact
• Cross layer interaction
– Effective use of new radio technologies at higher layers
• Example: IEEE 802.11a (fastest on the market)
– Highest PHY rate: 54 Mbps, above MAC: less than 30 Mbps
– With RTS/CTS: less than 24 Mbps even under no contention
IEEE 802.11a without RTS/CTS (1472 byte packets)
IEEE 802.11a with RTS/CTS (1472 byte packets)
36
30
24
Theoretical
18
Intel (BSS)
QualNet
12
6
0
Effective Throughput [Mbps]
Effective Throughput [Mbps]
36
30
24
Theoretical
18
Intel (BSS)
QualNet
12
6
0
6
12
18
24
30
36
42
48
54
6
Physical Layer Data Rate [Mbps]
12
18
24
30
36
42
Physical Layer Data Rate [Mbps]
WMPG Workshop
August , 2005
48
54
Cross Layer Interaction
PDA application with
intermittent connectivity…
Application
Middleware Service
Cross Layer
Transport
Network
MAC
PHY
SDRs
Location service…
Smart Antenna
Others
UWB (802.11b…)
(SISO/MIMO)
WMPG Workshop
August , 2005
Wireless aware TCP…
Multi-path routing…
Power conservation…
Adequate feedback from
PHY is one of the key
requirements to success
WHYNET Approach: Hybrid Testbed
A hybrid testbed for heterogeneous, wireless networking
• Heterogeneity in wireless technologies: MANET, wireless LAN, 3G
cellular, sensors, narrowband, wideband, UWB, …
• Hybrid testbed
– Combines realism of physical testing with scalability, flexibility and repeatability of
simulations (Zhou et al TOMACS, April 2004)
– Smooth transition from design to deployment
Next Generation
Radio Networks
MANETs
Distributed Simulation
Testbed
WLANs
Sensor
Networks
WMPG Workshop
August , 2005
WHYNET Testbed
Real Component
Next Generation
Radio Networks
Virtual Component
Synthetic
Synthetic
Applications
Applications
Streaming
Streaming
Video
Video
Conferencing
Conferencing
VoIP
VoIP
HTTP
HTTP
MANETs
Distributed Simulation
Testbed
Routing
Routing
Queuing
Queuing
Sensor
Networks
• Domain Policies
• Node Capabilities
• Service Discovery
• User/Node Auth.
• Roaming/Location Mgmt
Routing
Routing
Queuing
Queuing
• Domain Policies
• Node Capabilities
• Service Discovery
• User/Node Auth.
• Roaming/Location Mgmt
WLANs
MAC
MAC
Abstract
Abstract
PHY
PHY
• Flexible & realistic evaluation
framework
• Smooth transition from system
design to deployment
MAC
MAC
MAC
MAC
High Fidelity Measurement
High
Fidelity
Measurement
PHY
Based PHY
PHY
Based PHY
MAC
MAC
MAC
MAC
PHY
UWB PHY
PHY
UWB PHY MIMO SISO
MIMO SISO
Simulation
Physical
Repeatability,
Realism
controllability,
scalability
Emulation
Real applications &
network protocols
WMPG Workshop
August , 2005
Hybrid Testbed Case Studies: Overview
• XCP performance in wireless networks
 Collaboration between UCLA and HRL Labs
• Adaptive video streaming performance in ad hoc networks
• Using SCTP for transport layer Internet host mobility
support
 Collaboration between UCLA and Univ. of Delaware
• Bandwidth aggregation on multi-homed wireless hosts in
inter-working cellular and mesh networks
 Collaboration between UCLA and UCSD
• Evaluation of a distributed fairness algorithm (IFA) for
mesh networks in presence of real and diverse set of
applications
Adaptive Video Streaming Performance in Ad Hoc
Networks
• Evaluate adaptive video streaming performance in
presence of channel fading, congestion and node mobility
in ad hoc networks
• Use QStream as a representative adaptive media
application
 Optimizes two quantitative measures of video quality along
temporal and spatial dimensions
 Relies on TCP for rate control and drops low priority data during
congestion to maintain video quality and timeliness
Adaptive Video Streaming Performance in Ad Hoc
Networks
• Hybrid testbed usage – emulated wireless hosts running
QStream communicating with each other over a simulated
ad hoc network
• Observed complete lack of correlation between perceptual
and quantitative metrics, especially with node mobility
sQualnet Simulation Framework
•
•
Scalable framework as extension to Qualnet
Rich set of sensor network specific models
 Sensing and radio channels,
 MAC (S-MAC, T-MAC) and routing
(diffusion, DTN, tree) protocol
 Battery and power consumption models
•
•
Sensor Node
Functional Model
Hardware
Model
Application
CPU
Formal release of version 1.0
Adoption by growing user base
 UCLA projects: SOS Dynamic Sensor OS,
Ragobot, Ad hoc Distributed Control
Systems (ADCS), and Helimote Energy
Harvesting Aware Sensor Nets
 External academic users: UCSB, USC,
University of Bristol, University of
Missouri-Rolla, Iowa State, Nanjing
University, City University of Hong
Long…
 Industry users: SDRC, Boeing, HRL
Radio
Sensor
Stack
Network
Stack
ADC
(Sensor)
Actuator
Transducer
Wireless Channel
Sensor Channel
Joint work with Mani Srivastava
Battery
Model
Hybrid sQualnet Capabilities
•
Modeling of multitiered
heterogeneous sensor networks
 Field of motes (non-IP) with
backbone of microservers (IP)
•
Real-code emulation for motes
 Run unmodified SOS and TinyOS
code for motes in sQualnet
 Makes larger set of protocols
available, reduces debugging effort
•
(in progress) Hybrid simulation: mix
of simulated and real nodes
 Wireless channel emulation, sensor
channel emulation, actuator
emulation and application emulation,
Impact of (Real) Traffic Model
•
•
•
•
MANET traffic model (CBR)
 AODV and DSR perform equally
well
Sensor Network traffic model
 DSR performs 500 times better than
AODV
Lesson: Routing protocol performance
is highly dependent on traffic model.
Simulation: 1K nodes, 256B pkt, 1
pkt/sec for 15 min.
To summarize: High sensitivity of predicted network performance on
traffic distributions.
Testbed Challenges
• Scalability
 Provide use-sensitive projections: 1K MANET vs 10K node sensor
networks vs 1M node mixed network vs …
 Study layer-specific protocols vs. cross-layer effects
• Application-centricity
 Shift focus from protocol & device performance to impact on endend application (middleware) performance
 Impact on simultaneous support for multiple applications with
diverse QoS requirements
• Interoperability
Cost, robustness, efficiency … of dynamic switching among
networks