Review for TCP-LP: A Distributed Algorithm for Low Priority Data Transfer
Familiarity: Familiar with this area of research (2)
Recommendation: Likely accept (top 20% but not top 10%, significant contribution) (4)
Importance: Introduction of TCP-LP for low priority connections. TCP-LP is able to provide bandwidth for the low priority traffic without much impact on the regular TCP traffic.
1. A protocol to utilize the excess bandwidth in TCP networks without impacting regular TCP traffic.
2. A thorough simulation and modeling of the protocol to understand the behavior.
Strengths: A new approach; thorough analysis and evaluation.
Weaknesses: I am not totally convinced the usefulness of TCP-LP when multiple bottlenecks are present in the system. In particular, if one combines heterogeneity, bulk transfer and multi-hop web traffic.
Summary and comments: This is a novel idea and does not require changes in existing infrastructure as the changes are only at the end points. Although some of the simulations have shown the benefit of using TCP-LP, I am not sure if it will be able to react in real networks so that the TCP traffic is not affected.
Familiarity: Working in this area of research (3)
Recommendation: Definite accept (top 10%, excellent paper) (5)
Interest: The paper suggest a low priority TCP termed TCP-LP. The main idea is to throtle TCP-LP before it can take bandwidth that regular TCP can use. The result is a transport for low priority data that make use of bandwidth not available to the regular best-effort traffic.
Contribution: The paper shows how with changes at the client side only, TCP-LP can be implemented and comprehensive simulations demonstartes that it works very well. Additional formal analysis study the limitations of the scheme.
Strengths: It is a very good work. Iteresting idea, the algorithm is clear and simple to implement; comprehensive simulation study with additional analysis. very clear presentation.
Weaknesses: I realy believe this paper should be accepted.
Its main weaknesses are:
1) Though it can be implemented from the technical point of view, it will not in the current internet economy. What is the incentive? WIll I pay less than best effort? with high priority traffic things are clear - I'll pay more for what is important to me. Why should an ISP implement a pricing scheme that will lower its revenue?
2) The analysis in III-c is not impressive.
Familiarity:Working in this area of research (3)
Recommendation:Likely accept (top 20% but not top 10%, significant contribution) (4)
Importance : The paper presents a mechanism for reliable transmission of low priority data without significantly interfering with the TCP flows. This frees up network resources and hence, reduces response times of HTTP traffic.
These issues are certainly important. But it is not clear which applications fall in this category of low priority data transfer. Also it is not clear what the benefit to an application is if it uses low-priority transmission as opposed to regular TCP.
Contribution: Low priority end-to-end reliable delivery is an unchartered territory. However, most of the proposed mechanisms used to achieve that have been proposed before in other context.
Strengths: Low priority end-to-end reliable transmission is a new idea. Simulations under different loads and different network scenarios.
1. It is not clear what the application benefits by using TCP-LP as opposed to TCP. It appears to be a free community service.
2. The presentation of the results need more explanation. Especially what the two axes mean.
Summary and comments:
The paper proposes a mechanism for reliable end-to-end delivery of data using the left over bandwidth without significantly interfering with the ongoing TCP flows. It achieves that by early congestion notification which may be sent explicitly by the network or may be inferred based on one-way packet delays.
Comments to authors:
1. what if I transfer bulk data and I choose not to use TCP-LP? Or in other words, if I use TCP-LP will the network pay me something?
2. Equation 3 is unclear. Explain why the +C and -C terms are appearing there? give some details.
3. Figure 3 is confusing. remove the parantheses and write it in words rather than C code
4. Have the simulations used the reference mode (Fig 2(a)) or the realization 2(b)?
5. The presentation of the simulation results is hard to understand. Explain the normalization with some more detail when you first introduce it in Figure 11.