### Table 5.1: The properties of a sample overlay graph for various protocols.

2005

### Table 3. Average results of experiments of mapping two random graphs on the sphere to the plane. Calculating the overlay in the plane and mapping the resulting graph back to the sphere.

2002

### Table 1: Properties of three overlay topologies built for the sample graph showed in Figure 2. 2-SP denotes a 2-

2001

"... In PAGE 6: ... 2-SP denotes a 2- SPANNER, described in section 4. A sample graph is shown in Figure 2, with the metrics we compute displayed in Table1 . As expected, a minimum spanning tree (MST) offers the lowest cost at the expense of a high root RDP.... ..."

Cited by 3

### Table 1: Connectivity: A 3-regular random graph and a 3-Araneola overlay has a connectivity of 3. The rest of the graphs have a connectivity of 1 or 0.

"... In PAGE 3: ... THE METRICS 4.1 Connectivity Table1 presents the connectivity of the different graphs. A k-regular random graph and a k-Araneola graph are almost always k connected [10, 8].... ..."

### Table 1: Connectivity: A 3-regular random graph and a 3-Araneola overlay has a connectivity of 3. The rest of the graphs have a connectivity of 1 or 0.

"... In PAGE 3: ... THE METRICS 4.1 Connectivity Table1 presents the connectivity of the different graphs. A k-regular random graph and a k-Araneola graph are almost always k connected [10, 8].... ..."

### Table 2. Performance Results of the Prefix Overlay Transducer

"... In PAGE 9: ... We have tested the transducers on a 2 MB (300000 token) subset of the Reuters corpus that was tagged with part-of-speech tags and morphological features with a statistical tagger. Table2 shows numbers for two trivial graphs that contain only one transition, matching exactly once (trivial1) or the most frequent word the (trivial2). The result shows the maximum overhead for using the overlay matcher, as the original control state must be built at each match.... ..."

Cited by 1

### Table 1. Latencies and bandwidth of the overlay networks.

"... In PAGE 6: ... 5.4 Network Benchmark Results on the Overlay Network Testbeds Table1 shows latencies and bandwidths of the site-to-site VPN and emulated two-site testbeds. We conducted the ping-pong tests over the OpenVPN and PacketiX gateway machines as well as the NIST Net router.... In PAGE 6: ... The graph named Native in each figure shows the performance when no gateway or route is interposed between the two switches. As shown in Table1 , the latency overhead by OpenVPN and PacketiX was 0.11ms and 0.... ..."

### Table 3: The join cost: A 3-Araneola overlay achieves the lowest join cost. 3-Araneola overlay achieve the best results in term of all four metrics. Moreover, using such overlays eliminates the main drawback due to which unstructured overlays were abandoned, namely the search inefficiency. In fact, with such overlays, one can reach up to 20% of the nodes with almost perfect search efficiency. As opposed to a 3-regular random graph, a 3-Araneola over- lay supports dynamic user behavior. In such an overlay, each single join or leave operation is handled locally, and incurs the sending of only 9 messages on average (or O(log N) mes- sages in the absence of a membership service). Therefore, we conclude that a 3-Araneola overlay is an excellent solution for a flooding-based peer-to-peer lookup system.

"... In PAGE 5: ... Note that a scalable membership ser- vice amortizes the logarithmic cost of knowing a random node by aggregating membership information, and hence it is more efficient than a random walk for retrieving random node iden- tities. Table3 shows the join cost for each graph in both cases. In a 3-Araneola overlay, a join operation requires sending 3k = 9 messages, assuming the existence of a membership service.... ..."

### Table 3: The join cost: A 3-Araneola overlay achieves the lowest join cost. 3-Araneola overlay achieve the best results in term of all four metrics. Moreover, using such overlays eliminates the main drawback due to which unstructured overlays were abandoned, namely the search inef ciency. In fact, with such overlays, one can reach up to 20% of the nodes with almost perfect search ef ciency. As opposed to a 3-regular random graph, a 3-Araneola over- lay supports dynamic user behavior. In such an overlay, each single join or leave operation is handled locally, and incurs the sending of only 9 messages on average (or O(log N) mes- sages in the absence of a membership service). Therefore, we conclude that a 3-Araneola overlay is an excellent solution for a ooding-based peer-to-peer lookup system.

"... In PAGE 5: ... Note that a scalable membership ser- vice amortizes the logarithmic cost of knowing a random node by aggregating membership information, and hence it is more ef cient than a random walk for retrieving random node iden- tities. Table3 shows the join cost for each graph in both cases. In a 3-Araneola overlay, a join operation requires sending 3k = 9 messages, assuming the existence of a membership service.... ..."