In this section, we will investigation the literatures related with the peer-to-peer overlay network and mismatch problem in overlay network.
In general, peer-to-peer overlay network can be divided into two basic categories including unstructured peer-to-peer overlay network and structured peer-to-peer overlay network. The
Fasttrack/KaZaA[9], BitTorrent[10] etc.. The hosts in these overlay network are connected in a random and distributed way. In order to query the data pieces in overlay network, these overlay network use flooding, random walks or expanding-rich TTL approaches to query data pieces. Figure 2 and Figure 3 shows the search example with respect to Gnutella[6] and Freenet[7]. The query message is sent to other peers to search the needed data pieces and the query message is also forwarded by these intermediate hosts to enhance the query scope. But the problem that message will be sent and reply in a long period will be produced. Therefore, another architecture Fasttrack/KaZaA[10] is proposed to enhance the query performance. In this architecture, a super node is proposed to index the data pieces and the nodes hold the data pieces belong to the super node. The hosts in overlay network are organized as the hierarchy architecture. Figure 4 shows the example of Fasttrack[8].
Figure 2 : An Example of Query Message in Gnutella[6]
Figure 3 : An Example of Query Message in Freenet[7]
Figure 4 : An Example of Fasttrack[8]
Another kind of peer-to-peer overlay network is structured peer-to-peer overlay network. The famous are CAN[1], Chord[2], Pastry[3] and Tapestry[4]. These architectures are based on Distributed Hash Table(DHT) mechanism to allocate the hosts in the overlay network. DHT assign key to the data and compute a value for the key. The (key, value) pair is used for retrieving and locating the data item on a peer. CAN[1] is the first architecture of peer-to-peer overlay network. The hosts are located in a geographic way. The overall hosts’ spaces are divided into d-dimension Cartesian coordinated spaces. Each host in overlay network belongs to on a distinct zone in the overlay network. Figure 5 shows the example of the CAN[1].
Figure 5 : An Example of CAN[1]
Chord[2] is another famous structured peer-to-peer overlay network. Chord[2] organizes the hosts in a ring structure. Each host maintains a finger table as the routing table and the routing path will be according to the finger table. The query message will be transmitted in a clockwise way until the data piece is found or not. Figure 6 shows the example of the Chord[2].
Figure 6 : An Example of Chord[2]
The above peer-to-peer overlay networks all have mismatch problem because they do not consider the underlying network topology in constructing the overlay network. Liu et al.
propose several solutions to resolve the mismatch problem for unstructured pee-to-peer overlay network. In 2004, Liu et al.[12] propose a location-ware topology matching(LTM) technique to solve mismatch problem in unstructured peer-to-peer overlay network. In LTM host does not require global topology of peer-to-peer overlay network to optimize the overlay network structure. LTM issue a detector to detect the delay information in a constrained range(hops) and hosts collect these information to estimate the optimized overlay network structure. At the same time, Liu et al.[13] also propose another mechanism Adaptive Connection Establishment(ACE) algorithm to resolve the mismatch problem. Hosts collect the delay information by send the probing message and then calculate the minimum spanning tree as the optimized overlay structure according to the collecting delay information. Based on the optimized overlay structure, the host will probe other hosts to find the other closed host and try to establish overlay connection with the better host. Another mechanism scalable bipartite overlay scheme(SBO) is proposed by Liu et al.[31] to resolve the mismatch problem and that is similar with the ACE algorithm. But SBO divide the hosts into two types, one is responsible for collecting delay information and the other is responsible for calculate the optimized overlay structure. Therefore, overall performance of algorithm of finding optimized overlay network is improved.
Xin Yan Zhang et al.[22] propsoe mOverlay to resolve mismatch problem also. They take the locality of the hosts, i.e. distance, into account to construct the overlay network by using dynamic landmark. They introduce the group concept that hosts in group have same distance with group’s neighbors. The group’s neighbors are the dynamic landmark node to find the minimum distance.
Tongqing Qiu et al.[23] propose a generic approach to construct the topology-aware overlay network and they also use landmark as the basic scheme. By using the information from the landmark, the two hosts will be swapped to have better performance in overlay network. The decision for swapping is based on the calculation of delay information before swapping and after swapping. If the performance before swapping is better than after swapping, the two hosts will not swap to exchange hosts’ information. Otherwise, they swapped to have better overlay network topology.
Guangtao Xue et al.[24] propose a two hierarchy architecture to construct overlay network topology to resolve mismatch problem. The hosts in lower hierarchy are closest hosts. If the
closed host can not be found in the lower hierarchy, they will search for high layer hierarchy to find the closed hosts. Based on the two hierarchy architecture, the locality of the hosts in overlay network can be realized to solve the mismatch problem.
Guoqiang Zhang et al.[25] propose a simple approach to solve mismatch problem by collecting global information from BGP table in Internet. The global information reveals the global information with respect to the hosts in the overlay network. Therefore, the topology-aware topology can be constructed in a simple way.
Zhichen Xu et al.[15] combine landmark clustering and RTT measurement have accuracy information to construct topology-aware overlay network topology. Kai Shen et al.[27] use landmark hierarchy information as the basis to construct self-organized DHT protocol to have better structured peer-to-peer overlay network topology. Sylvia Rantnasamy et al.[28] use landmark information to calculate the network latency and they present binning scheme to divide the hosts into different clusters based the landmark information. Therefore, the hosts in the same cluster will have short distance and the better overlay structure is established. Shansi Ren et al.[16] use the TTL-k flooding and RTT measurement to have the latency information within k hops. Based on the information, they will adaptive estimate the overlay network topology to solve mismatch problem. Demetrios Zeinalipour-Yazti et al.[30] propsoe domain name server ordering scheme to solve mismatch problem by helping with domain name server.
The hosts have same domain name server ordering will be closed with respect to the underlying network topology.
The basic solution of the above literatures is to gather the topology status of the hosts in overlay network and then produces the optimized topology. The topology status is defined as the delay information between hosts. But in our opinion, the delay information just reveals the performance of the association path between hosts in overlay network and can not represent the actual topology of the overlay network. Therefore, our proposed method does not consider the delay information only and we also take the underlying network hop information into account.