This paper introduced the basic mechanism of content distribution network (CDN) and how to measure the performance of CDN. The authors selected several CDN providers and compared the performance of their services. Finally future work and conclusions are given.

The strength of this paper:
1. It is a good paper to both beginners and researchers of CDN. It gives the basic concept of CDN and also gives the method to measure it. So beginners can learn a lot from this paper and researchers can get some useful information about the performance of CDN.

2. Many CDN providers's data are used and compared. Although CDN is a new industry, the authors find a lot of hot CDN providers and give a detailed service performance of them. So we can have a general view about the performance of CDN after reading this paper.

3. A lot of tables and figures are given in this paper. So it is easy for us to read and understand the content of the paper.

The weakness of this paper:
1. We need to know more about CDN from their data. For example, What kind of CDN has better performance? Although they give some reuslts to show the response times of different CDNs, we can not tell the reason of the results. Is it because they use different mechanism such as URl rewriting or DNS redirection? or because some CDN providers have more servers in the world or the provider has a closer server to the authors experiment spot? So more research may be done to give answers to this question. This can both benifit the CDN provider to adjust their method and the customer to know how to select CDN not only based on the response time.

2. As mentioned in 1, different location may experience different response time even when you get same object from the same CDN provider. So to do this experiment, it seems better to have distributed experiment locations that send the request simultaneously, and then compare the average response time from all these locations.

3. in the table VIII, the serial-1.1 of 54 images has smaller value than the parallel-1.0 of 54 images. This is an exception according to other results of the same table VIII. Intuitively, parallel should have smaller time value than serial, but why this happened? No analysis is given in the paper.

4. In the part that they tested the performance with fixed IP address and new IP address with DNS lookup, they only use same objects and download them again and again. This is not the case in reality. It is possible that different objects are in different CDN servers. So the DNS lookup is quite necessary for different objects. From their results, we can only tell that to get same object, fixed IP is better than the DNS lookup process. But for different objects each time, this may not be the case. So to show the performance, they need do more tests with different objects every time and to see if the fixed IP method still work for that situation.

5. They used some NLANR logs for getting the nature of HTTP-requested CDN content. But they didn't say how they can get this information from the logs. It seems they catagorized data according to the name apppeared in the URL. What if the CDN provider uses IP address to send the response so the request only contains IP address instead of name?

6. As they mentioned, they haven't handle the streaming media content. This may have effect on their experiment results. For example, it is possible that for large media content, different CDN providers will have different distributed storage management, so the response time may change accordingly.

Genarally speaking, this paper is a good paper talking about the CDN and how to measure it, and some useful experiment results are also given. We can learn a lot from this paper. Still more research can be done based on their experiment.