MobiCom 2006 - Day 1

300 paper submissions 32 countries

Best student paper award: Measurement Study of Vehicle Access (MIT paper)

Keynote

Mobile Broadband Evolution - from Controversy to Convergence

Andrew Viterbi

Twin Engines of IT Growth

Internet (1 Billion)
Mobile phones (2 Billion)

Wireless and Internet Penetration

No America: 70/70
Europe: 75/40
Asia: 18/9
Global: 30/15

Analog entrenchment in N.A. made NA-TDMA/GSM/CDMA slow to take off
Asian experience is what finally drove CDMA's success
Today, approximately equal division between CDMA and GSM
GSM
- 2 billion worldwide
- early start
CDMA
- 400 million
- superior technology
3G
- CDMA2000-EvDO
- DoCoMo and EU: W-CDMA (GSM compatible)
- China: Some form of CDMA
3G Motivation
- NA/Asian: high speed data
- EU auctioned spectrum at high cost
- NA geared service to business and professional
- EU carriers target consumer market
4G: OFDMA/MIMO (WiMax)
Wireless first exceeded Wireline in 2000 (600 Million)
Today Wireless is about 3x Wireline
Wireline broadband is at about 1 billion
Wireless broadband is approach 100 million
Applications
- Location-based
- Built-in camera
- Quality Audio
- TV/Radio broadcast
Which of these apps benefit from broadband wireless?

Session 1

Computationally Efficient Scheduling

Critical Issue: Maximize capacity of wireless backbond
CSMA/CA is too conservative
- Possible solution: consider true interference and use spatial-reuser approach (S-TDMA)
Differentiate b/t physical and protocol interference
Formulated problem: given comm graph, find minimmum-len schedule satisfying demands such that in each time slot the SINR threshold is maintained at every participant (according to physical interface model)
Future work
- Integrating physical scheduling with physical
- Distribution scheduling with knowledge of pairwise signal strength

A Coordinate-Based Approach for Exploiting Temporal-Spatial Diversity in Wireless Mesh Networks

Using Soekris boards
Characterize two types of interference (flow-level)
- Interflow
- Intraflow
Mitigating Interference
- Transmit power/topo control
- Carrier sens threshold
- Channel diversity
- Scheduling concurrent transmissions optimally
Intra-flow interference - can not be avoided in a single flow

(Can we use L3/4 flow predictions to improve link-layer scheduling?)

Issues to consider
- Topology discovery
- Transmission scheduling
- packet transmission interleaving
Topo Discovery
- RSS measurement based
- Computer SNR between two-hop neighbor nodes to get least interference nodes
Transmission Scheduling
- Select first packet
- Search up to N other packets going to non-interfering nodes
Packet Transmission
- Congested nodes have higher priority over neighbor nodes
Experiment
- Champaign-Urbana Community Wireless Network
- Concentrate on transporting downstream traffic
- Augmented NS-2 simulation
- Showed approximately 1Mbps improvement for loads over 4Mbps
Future Work
- Topology construction with various performance metrics
- More experiments in large scale networks
Audience questions
- Packet re-ordering, does it affect TCP fairness?
  - Didn't quite understand answer, but I think that the answer is that the queue re-ordering won't affect it too much
- Do you assume node synchronization
  - Synchronization not needed

Enabling Distributed Throughput Maximization in Wireless Mesh Networks - A Partitioning Approach

Consider wireless-mesh-networks for last-mile access
Consider time-slotted system with stochastic arrivals
- Only a subset of links can be activated simultaneously due to interference
Maximum Differential Backlog
- Applies to join routing and scheduling
Interesting concluding point - more interference can improve the performance of distributed algorithms.

On Accurate Measurement of Link Quality in Multi-hop Wireless Mesh Networks

Limitations to current systems
- BAP (Broadcast-based Active Probing)
  - Based on cheap broadcast
  - Easy to implement
  - BUT: Different PHY settings
  - BUT: Bidirectional measurements
- Unicast-based Active Probing
  - Same PHY settings
  - Unidirectional measurements
  - BUT: Capacity overhead
  - BUT: Blind to retransmissions at MAC layer
- Self-monitoring
  - Reduce probing overhead
  - Use unicast/unidirectional results
  - BUT: Require probing during idle
  - BUT: Blind to retransmission
New Approach: EAR (Efficient Accurate Link-Quality Monitor)
- exploits existing traffic by adaptive selection
- use unicast packets
- easy deployable at network layer and device driver layer for x-layer interaction
- inner ear and outer ear... haha
Results
- BAP 10.2% error
- EAR 1.6% error
Conclusion
- Asymmetry awareness improves end-to-end throughput two times
Future Works
- Measure QoS parameters
- Extend to mesh networks

Lunch Talk

Deborah Estrin

Sensor network research is not what we originally thought. Less about as many small devices that last as long as possible... that's too general.

More about hetereogeneity, sensing fidelity, and so on.

We are now at second-order research - not just doing same science in better ways, but allowing other fields to do NEW things.

One of the most interesting properties of sensor networks is their ability to self monitor and allow for instant feedback to the deployer.

Using cell phones as censors!

Session 2: Measurements

A Measurement Study of Vehicular Internet Access Using In Situ Wi-Fi

Networks

Answer the question
- What are the properties of "organically grown networks"
Opportunity
- Lots of idle broadband
- 65% have wireless
- What if users opened up their APs to share/sell bandwidth?
WiFi for Mobile Messaging
- Small cells
- Unplanned
- ~10 second connectivity
- Connection overhead
Data
- 232 days of normal driving with embedded PCs
- Scanpig scanner
- Associate -> Get IP -> Ping -> TCP upload

(Note: improving address acquisition for opportunistic?)

First end-to-end WiFi under normal driving conditions
Towards OpenWiFi Networks
- Currently legal issues (unauth access)
- Need more tech
  - Tiered accounting
  - rapid connection protocols
Questions
- James Scott - pessimistic because you only connect to one at a time.

Packet-Level Diversity - From Theory to Practice: An 802.11-based

experimental investigation.

Wireless seems a natural environment for taking advantage of diversity
Open-loop diversity
round-robin is close to optimal, and usually won't hurt

Long-Distance 802.11b links: Performance Measurements and Experience

What is effect of RSS on PER?
What is effect of packet size on TR and PER?
Is there a time correlation of errors?
Max app throughput?
Effect of interference?
Effect of weather?
Effect of MAC ACK timeouts on throughput?

Analysis and Implications of Student Contact Patterns Derived from Campus

Schedules

Fundamental question: how are people connected?
Students are connected if in same place at same time
Out of class contacts will significantly speed up dynamics
Need measurement based models for out of class contacts
Can this data be used to model real world contact?

Poster Session

Some posters I want to explore more:

Fine grained power control is not a good idea, because signal fluctuations

make close power levels essentially similar

Vivek Shrivasava at UW Madison
Haggle demo was running. Code is available online. Download it and play

around!

Channel Assignment based on traffic characteristics
- Aditya Akella at UW Madison
- UT Austin researchers too
Purdue: MeshCache for caching in Mesh Networks
Only cached along fetch paths, what about more aggressive/predictive caching

for larger networks?

Waterloo had a demo of OCMP running... posted pictures to Flickr
Someone at the School of Mines had a incremental improvement on GPSR