2008-10-02 Readings - Wireless Networks in the Real World

MODELING WIRELESS LINKS FOR TRANSPORT PROTOCOLS

Andrei Gurtov, Sally Floyd

Summary:

Talks about how to model real world wireless links. Critical topic because modeling and simulation is precursor to any extensive and/or expensive field tests. Includes discussion about essential aspects of models, examples of bad models in literature, modeling various link characteristics, modeling queue management, effect on mobility, and wireless link vs. transport protocol issues.

Background:

Needs to know basic wireless 802.11 type knowledge.

Discussions and Criticisms:

Paper talks well about science. More helpful would be a operational checklist of common modeling pitfalls, and list of how to model each aspect of network. This way somebody wanting to use wireless link models can quickly reference the essential results of the work. For my own future reference, I make the list below.

Common modeling pitfalls (may apply for wired as well)

Unrealistic models:
- TCP with unrealistically high loss rates.
- Evaluating active queue management by focusing on TCP start-up.
- Assuming regular inter-packet delays.
- Constant TCP retransmission timer.
- Using deprecated TCP versions.
- Modeling WLAN as duplex.

Modeling a subset of parameter space:
- Evaluate using a single flow.
- Assume a particular link is bottleneck and not try the opposite assumption.
- Assume reserves transmission.

Overly realistic models:
- Focus on transient implementation flaws.

Lack of reproducibility (my routine complaint against a lot of methodological descriptions, far less emphasized today as it was in the early days of modern science):
- Not giving enough modeling details.

Also, list of wireless link aspects being modeled:

- Error loss & corruption.
- Delay variation.
- Packet reordering.
- On-demand resource allocation.
- Bandwidth variation.
- Asymmetry in bandwidth/latency.
- Queue management.
- Mobility.


ARCHITECTURE AND EVALUATION OF AN UNPLANNED 802.11B MESH NETWORK

John Bicket, Daniel Aguayo, Sanjit Biswas, Robert Morris.

Summary:

Talks about the architecture and realization of the Roofnet system. Key aspects of the system include totally unplanned - volunteers sign up and get antenna kit, omnidirectional antennas, use multi-hop routing, with estimated transmission time (ETT) metric - derived from estimated transmission count (ETX), which is presented in a separate paper.

Background:

Probably should read the ETX paper before this one. Also 802.11b knowledge. Maybe something about the different mesh routing algorithms would also help.

Discussions & Criticisms:

Pretty thorough performance characterization. However most of the results have no baseline for comparison - ok it's good, but HOW good is it? Not clear what should be the baseline of comparison either. Some kind of planned network? Directional antenna? Single hop to gateway? Some mixture of everything?

Tables 4 & 5 pretty valuable - comparison of multi-hop and single-hop networks.

Two key insights:
- Transmissions on multi-hops will interfere with each other.
- 802.11 RTS/CTS is BROKEN. They found RTS/CTS doesn't do what it's supposed to do, and introduces overhead. The evaluation should be repeated for other more general 802.11 settings. If results are reproduced, then we need to give general advice that RTS/CTS should be turnned off.

One evaluation that I think they missed - evaluate network performance under heavy load/utilization. Their network is utilized around 70% - for the Internet gateway i.e. - can we push it up to heavier loads? Also, scalability of gateway - all traffic funnel through gateway, so gateway = bottleneck? This will be apparent only if the network more heavily loaded.