Paper review: A comparison of mechanisms for improving TCP performance over wireless links

Reviewer: Kevin Hofstra

1. How can we improve the end-to end performance of TCP over lossy, wireless hops? Must the connection be split into a lossy and a reliable section? Should the packet loss be detected in the link layer or the transport layer?
2. A comparative analysis of several distinct techniques to improve end-to-end performance in lossy network segments. A presentation of the different methods for improvement followed by an actual study of their performance in practice.

A. The 3 techniques for increasing end to end throughput and goodput are:

i. A reliable Link Layer protocol that uses knowledge of TCP. (example: LL-TCP-AWARE)

ii. A split-connection approach to divide the lossy and reliable sections. (example: SPLIT-SMART based selective acknowledgement)

iii. End to end schemes (example: E2E-IETF-SACK)

2. The performance of the TCP-aware link-layer schemes is about 1.75-2 times better than E2E-SMART and about 9 times better than TCP-RENO. It also has noticeably better goodput than SPLIT-START.
3. The objective of the new wireless protocols is to provide a mechanism by which the transport protocol can be made aware of losses unrelated to network congestion and react appropriately to such losses.

4. Critique the main contribution
• Significance- 2 The article is more of an evaluation of the current implementations than an actual discovery. There evidence is somewhat conclusive, but should be considered more of a review than a discovery.
• Convincing- 3 Each of the implementations are well explained and they do present some actual network data. But the data they present is over a small spectrum and not necessarily representative in all cases. Many of their claims lack scientific proof or adequate data.
5. System researchers and builders should recognize that the lossy wireless links in an end to end connection do not necessarily have to be segregated from a normal reliable section. They must use different acknowledgements or perform reliability at the link-layer, but do not have to be split. As we continue to add more heterogeneous types of network segments to the internet there is increasing demand to provide seamless integration without making changes to the internet as a whole.