Reviewer: Maxim Podlesny
Date: 9-8-2005
How would you rate this paper, relative to others we have read? top 25%, but not top 10%
How would you rate your knowledge of the topic of this paper? familiar, but not expert
What problem or issue does the paper address? Why is it important?
The problem that is considered in the paper is of accurate measuring of packet loss. The problem of packet loss is very actual in protocol design as well as in the developing Internet-based applications.
What are the main contributions of the paper and why are they important?
The main contribution of the paper is of designing a new algorithm for packet loss measurement and of developing a tool that implements the algorithm. The algorithm is based on several assumptions about accuracy of packet loss showing up and the rule of sending probes.
How significant are these contributions relative to previous work?
Authors have shown that the method of packet loss measuring based on Poisson-modulated probes does not have a good accuracy for all the types of traffic. The accuracy of a new method is better than of existing ones.
Give detailed comments justifying your view of the paper.
The paper has a good structure as well as explained main ideas in a detailed form. The main theoretical concepts concerning packet loss were described in very good manner. The accuracy of the method is very good for different types of traffic and lengths of loss episodes. But it is not clear why for reporting of loss episodes the length of a probe was between 1 and 10 packets. Probably, taking an interval between probes less than 10 milliseconds would be better. Concerning figure 7, it is clear what accuracy is for different lengths of a probe in case of constant-bit rate traffic. But it is not clear what would happen if a probe has a length equal to 15 packets in case of infinite TCP traffic. Authors said that longer probes began to have a serious impact on the queuing dynamics during loss episode, but no graph showing that when the lenghth is more than 10 packets is provided. In addition, there is no explanation about what happens when the length of a probe is equal to 2 in the same case. Why does the inflection of curve take place for the length of a probe equal to 2? Probably, the curve has a different behaviour in case of other parameters of the experiments made for accurate reporting. It would also be interesting to know if the accuracy of the method can be improved by taking somehow into consideration 'history', i.e. previous time slots in the assumptions about the basic algorithm. In the end, I did not find any serious disadvantage in the paper.