Paper Review: A comparison of mechanisms for improving TCP performance over wireless links
Reviewer: Kenneth Chin
This paper classifies wireless protocols into three main categories: 1) end-to-end; 2) link-layer; 3) split-connection. The purpose of this paper is to investigate the performance improvement from different schemes within their own categories, and concludes that, after inter-comparison, the link-layer scheme best suits the wireless network.
The following are the main ideas from this paper:
Basically, a wireless network is not entirely wireless in the sense that it is coupled with a wired network. It is commonly understood that TCP works quite well in wired network, but it has a poor performance when it comes to wireless network. The major reason for the lousy performance is due to the fundamental assumptions that all packet losses are due to congestion. However, due to relatively much higher loss rate in wireless network, unnecessary window size reduction leads to poor throughput. Also, the cumulative acknowledgment scheme used in the current TCP is also a big problem. In view of these issues, people come up with selective acknowledgment (SACK) and SMART schemes in replacement of the cumulative acknowledgment scheme. Furthermore, an explicit lost notification (ELN) mechanism is introduced into TCP to boost performance.
Another way to improve performance is to shield the sender from wireless loss by doing things in the link-layer. In this way, upper level will be handling end-to-end congestion. The base station cache packets and retransmit lost packet to the mobile hosts and suppress duplicate acknowledgements to the sender. All these are done in the link-layer. The author argues that the link-layer approach is the best for wireless network. However, the sender is not completely shielded from wireless loss.
The main idea is to shield the sender completely from wireless loss. In this approach, the wired network still uses TCP while the wireless network uses another protocol. According to the author, this approach might not be as good as the link-layer approach because it violates the end-to-end properties and also acknowledgments might arrive at the sender before the packet arrives at the receiver. What's more is that the sender in the wired network is stalled when the wireless network cannot accommodate the traffic.
This paper did have some contribution, so it should get a 4th grade.
- split-connection is not necessarily helpful
- selective acknowledgement (SACK) and explicit loss notification (ELN) are very important.