In this paper Savage, Cardwell, Wetherall, and Anderson describe three exploits of common TCP stack implementations where a misbehaving receiver can force the sender to send at a faster rate. As we have seen, all of the problems come from rather informal descriptions of how TCP should behave (e.g. byte versus segment granularity and so on). In addition to the exploits the authors also propose fixes, some of which solve the problem but require a modification to the TCP header, others that mitigate the effect a misbehaving receiver without a modification. To proof their theory the authors implemented the exploits in the Daytona TCP stack.

The paper was short, precise, and easy to understand. Unlike for other papers no question about what the authors covered was asked (either because of the simplicity of the topic or the good explanation of the authors). Supporting figures helped to visualize the exploits.

During the discussion the question was raised if those exploits are also applicable for larger file transfers. The paper did not mention that detail and presented only experiments with a 60Kb file but most students believe that there is no problem. Another question was how much impact the wide use of that stack has on the Internet. The authors claim that it could cause congestion collapses and therefore should be fixed as soon as possible. The class did not take it that seriously. Last but not least Dr. Turner mentioned that those ideas/exploits are also applicable to other protocols.

At the end of the discussion the paper was rated quite well. The upper and middle third of our scale did receive a roughly equal number of votes and I believe no vote was given to the bottom third. A main reason for not rating it as one of the top papers was the simplicity of the research.

Christoph Jechlitschek