Cloud computing is a very timely and important topic, and there is current debate on whether it should be centralized or decentralized. The authors try to focus on providing an architecture to bridge the gap between decentralized cloud computing infrastructures. Apparently, from a networking point of view, centralized networks have various problems, which include causing bottlenecks and poor fault tolerance. However, a great amount of effort has been put into developing warehouse-scale data centers to host cloud computing (for example, Google data centers and Microsoft data centers). The future of decentralized clouds, namely distributed data centers, has been questioned in the industry.

The authors propose the architecture of inter-cloud communication with three layers: access layer, distribution layer, and organization layer. Each layer has its own function. For example, the distribution layer has Dnodes to exchange information between clouds. The organization layer has Onodes to secure the system. In the analytic model, the authors provide a nice study of their model, showing such information as how many nodes will be involved in a cloud, and how many Dnodes/Onodes will be needed. However, there remains an important concern: contemporary cloud computing infrastructures are very heterogeneous, and cooperation among these infrastructures is hard. For example, the protocols of a data center from application layer to even networking layer could be very different among various cloud computing environments. After applying the inter-cloud management architecture, these heterogeneous cloud computing environments may still not be able to communicate because of the different underlying protocols.

In sections 7 and 8, the authors present a comprehensive simulation study of their inter-cloud architecture. Their evaluation results seem to be interesting. Many scenarios are covered, such as many nodes among many clouds, bandwidth, messaging, broadcasting, and so on. The architecture seems to be feasible and lightweight. Again, their setup is somehow a homogeneous environment (the same cloud with the same set of machines), which is different than a real-world setup where clouds have heterogeneous protocols and machines.

Overall, I think this paper is an interesting study about inter-cloud communication, but the importance of this study is somewhat limited.