Modeling wireless Links for Transport Protocols

Wireless links face issues like variable bandwidth, corruption, variable latency etc. This affects the performance of TCP because TCP assumes packet losses are due to congestion. Link layer protocols like FEC/ARQ have been proposed to improve performance but they suffer from drawbacks like large link-level queues.

This paper proposes models that should be used for evaluating transport protocols over wireless links. Wireless links include Cellular, WLAN and satellite. The authors say that the traffic model can have significant effect on the simulation results. It is often the case that if we do not model all types of traffic in the network then the evaluation of the protocol will not be complete. Transport protocols are generally evaluated over throughput, delay, fairness and dynamics. In case of wireless networks it makes sense to see how much useful data got through with respect to all the delivered data. The authors term this as "goodput".

Models used should be realistic, explore all the parameters, do not use something overly realistic and should be available to be used by others. The paper gives very good examples related to these features.

Modeling of wireless links include:

1. Error loss and corruption should be modeled by dropping packets at a per-packet, per-bit or time-based loss probability. Losses due to data corruption should be modeled at the end of the link while handover losses should be modeled before the link.
2. Delay variation can be modeled by suspending the data transmission on a simulated link.
   
3. Packet reordering requires either building a queue and swapping packets or delaying one packet for a given time.
4. On demand resource allocation can use a model where additional delay is introduced when a packet arrives to a queue that has been empty for a longer time.
   
5. Bandwidth variation can be modeled by having periods of low and high bandwidth.
6. Asymmetry is modeled by having different settings for uplink and downlink.
7. Queue management
• WLAN/Cellular links: use drop-tail buffer with configurable maximum size in packets
• Satellite links: use RED
8. Effects of mobility have to be modeled based on whether underlying mechanisms of handover have to be evaluated or the effects of the handover need to b e evaluated.

The authors in the end discuss how bit errors in wireless links should be handled at the link level while reordering of packets be handled at the transport level. Transport protocols can be made robust to address delay variations while link-level protocols can be changed to hide delay variations from the transport payer.

The paper was interesting to read because it explained different environments that wireless communication can face and how the models should take these into consideration.

The paper says that currently wireless links at both ends is not common. Is this true today as well? I am not sure what the authors mean. Do they mean that complete communication uses wireless links or that both the machines communicating are not using wireless links?