Self-Clustering Sample Clauses

Self-Clustering. The Internet as a whole provides sporadic and unstable connectivity, e.g., web users frequently experience disconnects and server failures. The instability can occur because of congestion, equipment failures or lossy links. It can also take place as a result of denial-of-service attacks, worms and viruses. It is often the case that an unstable network component (router or link) tends to have multiple failures. In other words, an isolated, “once-in-a-blue-moon” type of failure is uncommon. Repeated failures typically complicate protocol implementation. However, oddly enough, TGDH not only survives but also benefits from repeated failures. Similar to other tree-based key management schemes (e.g., [36, 38, 24]) the key tree in TGDH is logical: group members are leaves in a tree and internal nodes are logical. The initial placement of members (as tree leaves) is not dependent on their relative physical location. Therefore, members physically close to each other might not be neighbors in a key tree. When a partition occurs, all members in the same physical group fragment form a new key tree and a new group. The partition protocol may cost as many as rounds. Then, when the partition heals, the previously separate groups are merged into a single key tree, however, they are still clustered along the lines of the partition. If another partition happens on the same link, the partitioned members are not scattered across the key tree any longer. Therefore, any subsequent partition on the same link will take only one round to complete. This property is especially important in high-delay wide area networks since clustering lowers the number of communication rounds as well as the number of modular exponentiations, in many cases. First Partitions on a weak link L All merges All other partitions on the weak link L Figure 13 shows an extreme example of self-clustering. Suppose that a group has sixteen members numbered through where white odd-numbered nodes are located in one physical cluster (e.g., a LAN) and shaded even- numbered nodes in another. The two partitions are connected via an unstable link . If fails and a partition occurs, it takes three rounds to complete the partition protocol. It can be clearly seen that each group forms a cluster after the partition. When comes up and the partition heals (i.e., a merge occurs), two rounds are needed to complete the merge protocol. Subsequently, all partitions on link will require only one round and all merges – two rounds.