This paper talks about several algorithms of searching and replication in unstructured peer-to-peer network. Compared with the flooding method, they talk about expending ring and random walks for searching in the peer-to-peer network. For replication, they studied the uniform, proportional and square-root repliaciton methods. Some simulation results are given for these methods.

The strength of this paper:
1. Real peep-to-peer data is used in the simulation, so the results are more practical and convincive. For example, the Gnutella graph which is obtained in OCt 2000 was used as one of the network topologies in the study.

2. A lot of comparisons are given in the paper, so we can learn better by these results. For example, Figure 3 in the paper shows different metrics vs. TTL value, and for each metric, 4 topologies are shown. So we can get a clear view of different metrics and different topologies after reading the figure.

3. The experiment data is abundant in the paper. We can see there is a lot of data in the Table 2 to Table 4. So we can see the authors' effort here.

The problems with this paper:
1. Although this paper is focused on Gnutella network and several algorithms are mentioned to compare with current flooding method in Gnutella, the authors didn't say why and how they get the idea of some methods, such as expending ring or random walks. And it is possible that there are several good searching mechanisms in other peer-to-peer network. What if we use these methods in the Gnutella environment? What is the performence compared to the authors' own idea?

2. Generally speaking, the structure of this paper is good, but it is still confusing in some places. For example, when talking about repliaction theory in Section 5, it seems that uniform, proportional and square-root are three mechanisms in the same level. But only square-root replication is in a bold style. This is kind of confusing. Do they want to show that this idea is optimal or superior?

3. Throughout the paper, they didn't tell how they implement their ideas or the detail of their simulation. So we can't tell for sure that if their simulation is correct or sensible. Although they give a lot of data in the paper, we are still not quite sure if we can trust these data.

4. We know in Gnutella users can input key words to search for files. So it is possible that they may want several files for one Query or they want as much information as possible? So how does the authors' checking idea work for this situation? They can only check that the requester has got some related files, but how could they know it is enough and they do not need to search further? And according to our observance of Gnutella network, a lot of Query messages are not for a certain file, they consist several key words.

5. In Section 4.3 they mentioned several principles of search in peer-to-peer network. For example, adaptive termination is very important or message duplication should be minimized. But these principles seem to be obvious goal of peep-to-peer network, people even without doing research should know this. They mentioned here as principles, but not much information is sent to us here.

6. Gnutella is only one of the unstructured peer-to-peer network, so to show the preformance of their methods, the authors should do more simulation or experiments for several unstructured peer-to-peer networks.

7. For the random replication idea, it seems to have optimal performance compared with other ideas. But how do decide the value k, that is how many nodes are selected to store the copy? It is important in the real implementation. No discussion is found in this paper. In fact, it shouldn't be hard to do so in the simulation. And with different k vs. number of total nodes in the system, we may get useful research results.

Generally speaking, this is a good paper. The authors give us a lot of data about performance of several algorithms for search and replication in Gnutella network. Still a lot of research can be done about this topic and related topics.