In this paper, the authors describe an infrastructure: Tapestry, which can send the messages to the nearest copy of an object. No centralized resources are needed and soft state is used to maintain the routing and location information. Several new ideas are introduced here, such as new ideas in routing and location, new idea for fault handling and dynamic algorithms for node insertion, location path recovery and introspective optimization. The simulation results of this infrastructure are also given in the end.

The strength of this paper:
1. A new infrastructure is introduced for routing and location in today's data network. Many algorithms are given in detail to show the special features of this infrastructure. So after reading, we have a clear view of what is this infrastructure.

2. Many new ideas based on the old system are mentioned in the paper. Such as using multiple roots, soft state maintainance. Some of these ideas seem to be reasonable for better performance. And the difference between the old methods and the new methods are mentioned, so we can know the authors' contribution to this infrastructure compared to related work.

The weakness of this paper:
1. The simulation results are not complete for comparison. For example, by using the multiple roots, the results are shown in figure 13 to demonstrate that the latency is reduced a lot with more roots. But we also know that by using multiple roots, we have overhead, such as more state or information packets and the node management of multiple roots compared to the single root. The paper did not give such overhead comparison, so we are not clear if the multiple roots idea is worth the cost of more bandwidth or computer resources.

2. Some ideas are not verified. For example, the paper mentioned that this infrastructure can support mobile hosts. But we did not see any simulation results show that this is true. The thing with soft state recovery is similar here. No any results show the mechanism can work. And we can know that to implement soft state mechanism, more packets are needed, so the overhead is quite important for the new system. Some analysis of the routing overhead are necessary for this paper.

3. They mentioned a lot of times that the Tapestry infrastructure is designed to ease creation of scalable, fault-tolerant applications. But after reading the whole paper, this conclusion is not quite convincible. From what has been described in the paper, we can not tell why this infrastructure can work better than some existing systems, why it has better fault-tolerant ability. To make this conclusion more believable, more experiments results of comparison between this new system and other existed ideas should be given.

This paper is only a introduction, not very convincible. Although several new ideas are mentioned in the paper, more results are still needed for other researchers to justify this system.