5. Peer-to-peer (P2P) networks PA191: Advanced Computer Networking I. Eva Hladká Slides by: Tomáš Rebok Faculty of Informatics Masaryk University Autumn 2018 Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 1/116 Lecture Overview Q Client-Server vs. Peer-to-Peer • Client-Server Systems • P2P Systems • Comparison Q Generic P2P Architecture • Overlays and Peer Discovery • Service/Resource Discovery Q Taxonomy of P2P Systems o Centralized P2P Systems • Decentralized P2P Systems • Hybrid P2P Systems Q Routing in P2P Networks • Introduction, Motivation • Routing in Unstructured P2P Networks • Routing in Structured P2P Networks • Routing in Hybrid P2P Networks Q Information Sources Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 2/116 Client-Server vs. Peer-to-Peer Lecture Overview I Q Client-Server vs. Peer-to-Peer • Client-Server Systems • P2P Systems • Comparison 0 Generic P2P Architecture • Overlays and Peer Discovery • Service/Resource Discovery Q Taxonomy of P2P Systems • Centralized P2P Systems • Decentralized P2P Systems • Hybrid P2P Systems O Routing in P2P Networks • Introduction, Motivation • Routing in Unstructured P2P Networks • Routing in Structured P2P Networks • Routing in Hybrid P2P Networks & Information Sources Distributed Applications I. • a distributed application consists of multiple software modules located on different computers • the modules interact with each other over a communication network connecting the different computers • the communication network is used for synchronisation and communication between the modules • it is possible that multiple users may use the application concurrently on different computers • to build a distributed application, it is necessary to decide: • how to place those software modules on the different computers in the network • how each software module discovers the other modules it needs to communicate with Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 4/116 Distributed Applications II. • two basic approaches: • Client-Server architecture • Peer-to-Peer (P2P) architecture • hybrids are possible and indeed useful Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 5/116 Client-Server vs. Peer-to-Peer Client-Server Systems Client-Server Architecture I. A client-server system comprises of two types of software modules: • server module • one centralized instance • but might be internally replicated for scaling purposes • passively listens for connections from clients • multiple client requests may be handled: • sequentially • concurrently (multithreaded servers) • by several replicated servers at different locations 9 pending clients' requests may be queued up • servers are assumed to be reliable, often running in a data centre (dedicated/virtualized hardware) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 6/ Client-Server vs. Peer-to-Peer Client-Server Systems Client-Server Architecture I. • client module o multiple distributed instances, possibly controlled by different users • actively initiates a connection to a server • no direct communication between clients • clients need to know the network address and port number of a server • service discovery is typically performed through client configuration • clients may be unreliable without affecting overall system stability • examples of client-server systems: • web server/web browsers • web server/client applications (web services) • SSH/Telnet/FTP server/clients • NFS/SMB server/clients Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 7/116 Client-Server vs. Peer-to-Peer P2P Systems P2P Architecture • a P2P system consists of many identical software modules (peers) running on different computers 9 peers communicate directly with each other • each peer is a server as well as a client: 9 provides services to other peers • requests services from other peers • unlike dedicated servers, peers tend to be unreliable 9 service discovery is more complicated since there are many servers continuously appearing and disappearing at different network locations • provide natural scalability due to multiple servers • can work without allocating dedicated server machinery Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 8/116 Client-Server vs. Peer-to-Peer P2P Systems Communication Structure Comparison Client-Server vs. Peer-to-Peer P2P Systems Peer-to-Peer Systems Peer-to-peer (P2P) systems are distributed systems consisting of interconnected nodes able to self-organize into network topologies with the purpose of sharing resources such as content, CPU cycles, storage and bandwidth, capable of adapting to failures and accommodating transient populations of nodes while maintaining acceptable connectivity and performance, without requiring the intermediation or support of a global centralized server or authority. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 10/116 Client-Server vs. Peer-to-Peer P2P Systems P2P Properties • Symmetric role • each participating node typically acts both as a server and as a client • however, in many designs this property is relaxed by the use of special peer roles ("super peers" or "relay peers") • Scalability • P2P systems can scale to thousands of nodes o the P2P protocols cannot require "all-to-all" communication or coordination • Heterogeneity • a P2P system is (usually) heterogeneous in terms of the hardware capacity of the nodes • Distributed control (Decentralization) • ideally, no centralized structures should exist in P2P systems • Dynamism • the topology of P2P systems may change very fast due to joining of new nodes or leaving existing ones • Resource sharing • each peer contributes system resources (computing power, data, bandwidth, presence, etc.) to the operation of the P2P system • Self-organization ^^^^^^^^^Jieorc^niTa^on^of the P2P svstem increases over time nsincr local knowledge Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 11/116 Client-Server vs. Peer-to-Peer P2P Systems P2P Applications Distributed databas Collaborative work environments Collaborative caching and storage Hlnstant Messaging Gaming Scientific computation Content sharing personal i enterprise Figure: P2P Applications. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 12 / Client-Server vs. Peer-to-Peer Comparison Client-Server vs. Peer-to-peer Comparison I. The systems can be compared from several points of view: • Ease of development • C-S is more established and familiar than P2P • C-S exhibits simple interaction patterns for clients and server, while P2P involves more complex interaction patterns between peers • Manageability 9 it is easier to maintain a centralized server in a C-S environment than keeping a track of and maintaining several distributed peers in a P2P system • Scalability • C-S scalability is limited by fixed server hardware, though scaling can be achieved through load balancing over multiple servers at increased cost • P2P is scalable by nature, since as the number of peers grows, so does the "server" capacity Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 13/116 Client-Server vs. Peer-to-Peer Comparison Client-Server vs. Peer-to-peer Comparison II. • Security • responsibility for the C-S security lies within the server, which is centrally hosted in a secure environment • responsibility for P2P security is distributed across peers in different administrative domains, some of which might be compromised • Reliability • the C-S's reliability is achieved through the use of multiple redundant servers (possibly hosted at different locations) with automatic fail-over, at additional cost • with P2P, resilience comes free of charge, since multiple peers are usually able to provide the same service in the case that some peers fail Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 14/116 Lecture Overview Q Client-Server vs. Peer-to-Peer • Client-Server Systems • P2P Systems • Comparison Q Generic P2P Architecture • Overlays and Peer Discovery • Service/Resource Discovery Q Taxonomy of P2P Systems • Centralized P2P Systems • Decentralized P2P Systems • Hybrid P2P Systems O Routing in P2P Networks • Introduction, Motivation • Routing in Unstructured P2P Networks • Routing in Structured P2P Networks • Routing in Hybrid P2P Networks & Information Sources Generic P2P Architecture P2P Architecture Application Layer O Middleware Layer o Base Overlay Layer o Underlying Network libraries exist that provide reusable P2P functionality (e.g. JXTA) some applications integrate all of the above (e.g., Gnutella, Bittorrent, etc.) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 16/116 P2P Architecture Base Overlay Layer I. • the base overlay layer is responsible for: • discovering new peers • maintaining the P2P overlay (virtual) network • forwarding messages between peers • the overlay network is a virtual network laid over the "physical" network (e.g. TCP/IP) • overlay network "wires" are implemented using underlying network facilities (e.g. TCP connections or UDP messages) • overlay network distance is measured in the number of hops from peer to peer • peers, that are distant in the physical network may be neighbours in the overlay network, and vice-versa • the performance of the P2P system is influenced by the structure of the overlay network Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 17/116 Generic P2P Architecture P2P Architecture Base Overlay Layer II. Figure: Overlay vs. Underlying Network. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 18/116 P2P Architecture Base Overlay Layer II. P2P Architecture Base Overlay Layer II. Generic P2P Architecture P2P Architecture Base Overlay Layer II. Eva Hladká (Fl MU) Figure: Overlay vs. Underlying Network. 5. P2P networks Autumn 2018 21/116 P2P Architecture Middleware Layer • the middleware layer facilitates P2P application development by hiding overlay and service discovery issues • it provides access to the services/resources provided by peers, and may be responsible for functions such as: • security: controlling access to services/ resources • service/resource discovery: searching and indexing services/resources distributed across peers • peer groups: coordinating peers that provide or consume a particular service/resource • may provide fault tolerance and persistent state o e.g., JXTA (Java P2P platform), BOINC, P2, Windows P2P Networking, Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 22/116 Generic P2P Architecture P2P Architecture Application Layer • the middleware services can be used to build complete applications: • file sharing - e.g., Napster, Gnutella, Kazaa, . .. • routing protocols • instant messaging, videoconferencing applications - e.g., Skype • distributed file systems • distributed backup systems • distributed computing - e.g., grid computing, SETKDHome, ... • and many many more. . . Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 23/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery • a P2P network is typically a "virtual" network overlaid on an existing network (e.g. the Internet) • the overlay is used for indexing and peer discovery and make the P2P system independent from the physical network topology • content is typically exchanged directly over the underlying IP network • a new peer needs to discover at least one existing peer in order to join a P2P network • network location information: IP address, listening port number, etc. • if no peers are found immediately, the new peer either • passively waits for new participants, or • proactively looks for potential new participants • it is hard to locate existing peers in a large network such as the Internet Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 24/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery Initial Peer Discovery I. Static configuration: 9 each peer is preconfigured with a list of the network locations (IP address and port number) of every other peer in the system 9 on startup (and possibly periodically) each peer attempts to connect to some other peers in its list, some of which may be running • due to the manual configuration, this is only suitable for P2P networks with a small number of peers which do not change frequently a can alternatively be used to initially contact a small number of "well-known" peers that are guaranteed to be online Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 25/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery Initial Peer Discovery II. Centralized directory: • each peer is preconfigured with the network location of a centralized server • each peer contacts the server on startup (and possibly periodically) to: • obtain an updated list of currently active peers • indicate to the server that it is active 9 most subsequent communications bypass the server, using the P2P overlay network to route messages instead • occasionally, other services are also provided by the server (e.g. a list of files hosted by each peer) o peers may go offline • cleanly, the peer's shutdown procedure contacts the server to remove it from the active peer list • without warning (crash, network or power failure), making the server's active peer list obsolete (it's necessary to use active peer list item expiry and periodic liveness checks) • usually, a peer only needs to connect to a few peers on the overlay network • the other members can be discovered by the member propagation techniques • centralized directory server is a single point of failure Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 26/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery Initial Peer Discovery III. Member Propagation Techniques with Initial Member Discovery: 9 in general, it is not necessary to discover all of the participating members in the network • in many cases, discovering a subset of the participating members is adequate • after discovering just one existing peer, information about the rest of the P2P network can be obtained from it 9 if each peer maintains a full member list —>* easy for any new peer to obtain a full member list from any other peer • alternatively, each peer can maintain a partial member list, replacing offline peers with new ones from neighbouring peers' lists Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 27/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery - Overlay Network Topology • intermediate peers in the overlay network forward messages between indirectly connected peers • the overlay topology significantly affects P2P system performance 9 two key properties determine the effectiveness of the overlay mesh: • Diameter: longest distance between any two peers (overlay hops or latency) • should be minimized • Average Degree: average number of links per peer (high AD increases message load, but improves fault tolerance) • should be kept at a moderate level • it is necessary to avoid linear formations and splits in the mesh • common topologies: • Random Mesh • Tiered • Ordered Lattice Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 28/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery - Overlay Network Topology Random Mesh • each peer discovers a number of other peers and attempts to connect to them indiscriminately • this (hopefully) results in a random structure with uniform degree • distant peers on underlying network could be overlay neighbours o solution: connect to peers with lowest latency • random mesh is suitable for linking a large number of peers with uniform resources and connectivity 9 search message flooding can easily be used to discover resources/services on other peers • but generates a lot of traffic Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 29/116 Generic P2P Architecture Overlays and Peer Discovery Random Mesh - Overlay Network Topology Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 30/ 116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery - Overlay Network Topology Tiered Structure • peers are ordered into tiers of a tree depending on their advertised resources and connectivity (e.g. Kazaa's nodes and supernodes, 2-tier) • tier 0 is the foundation tier containing (possibly well-known) reliable peers with adequate resources and message forwarding capacity • at each tier, every peer is linked to a number of peers of a lower tier and forwards messages up and down • poorly-resourced leaf peers only link to their 'super-peer' and do not forward other peers' messages; they are omitted from peer discovery • the system needs to recover from peers leaving abruptly and disrupting the tree structure • the hierarchy may be optimized to follow the underlying network's structure (e.g. P2P video streaming) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 31/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery - Overlay Network Topology Tiered Structure Tier 1 t 11 n í I ^1 I X Tier 2 /a- J .....í / ; Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 32/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery - Overlay Network Topology Ordered Lattice a in a two dimensional lattice, peers organize themselves in a rectangular grid: • each node maintains direct connections to 4 neighboring peers (except edge peers) • peers on opposite edges can also link to form a torus • can be extended to n dimensions • messages are routed parallel to the lattice axes a peer additions and deletions must be handled on the fly, possibly distorting the structure • insertions and deletions of nodes imply that different rows/columns have different numbers of members between themselves • peer coordinates in a multi-dimensional lattice may be used as a key to locate resources in content addressable networks (CAN) 9 sometimes also denoted as Distributed Hash Table (DHT) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 33/116 Generic P2P Architecture Overlays and Peer Discovery Overlays and Peer Discovery - Overlay Network Topology Ordered Lattice Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 34/116 Service/Resource Discovery • a peer must advertise its services to enable their discovery and subsequent use by other peers 9 e.g., in file sharing applications, the "service" is a shared file/block • service discovery is itself a service • centralized - a server is asked for service location o Napster, UDDI for web services • pure P2P - a request is flooded or hashed through the peers • flooding, overlay multicast, CAN/DHT • when a search message reaches a matching advertisement on a peer, the server's location is returned to the originator • actual service messages are either routed through the overlay or directly via underlying network by the application • can be optimized by caching advertisements/data (e.g. file/block) along search/return path on the overlay Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 35/116 Lecture Overview Q Client-Server vs. Peer-to-Peer • Client-Server Systems • P2P Systems • Comparison 0 Generic P2P Architecture • Overlays and Peer Discovery • Service/Resource Discovery Q Taxonomy of P2P Systems o Centralized P2P Systems • Decentralized P2P Systems • Hybrid P2P Systems O Routing in P2P Networks • Introduction, Motivation • Routing in Unstructured P2P Networks • Routing in Structured P2P Networks • Routing in Hybrid P2P Networks & Information Sources Taxonomy of P2P Systems I. Generally, P2P systems can be divided into two main categories: • centralized - one or more central servers are available providing various services • decentralized - no central servers are employed • they have to consider two main design issues: • the structure - flat (single tier) vs. hierarchical (multitier) • the overlay topology - unstructured vs. structured • besides these two, hybrid P2P systems also exist • they combine both centralized and decentralized approach to leverage the advantages of both architectures Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 37/116 Taxonomy of P2P Systems Taxonomy of P2P Systems II. P2P Architecture Structure erarchical Decentralized Unstructured Static configuration Re-configurable Topology Structured Precise Figure: A taxonomy of P2P systems. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 38/116 Taxonomy of P2P Systems Centralized P2P Systems Taxonomy of P2P Systems III. Centralized P2P Systems I. Centralized P2P Systems • combine the features of centralized (client-server) and decentralized systems • like a centralized system, there are one or more central servers, which help peers to locate their desired resources or act as task scheduler to coordinate actions among them • a peer sends messages to the central server to determine the addresses of peers that contain the desired resources • like a decentralized system, once a peer has its information/data, it can communicate directly with other peers • i.e., without going through the server anymore o drawbacks: • susceptible to malicious attacks and single point of failure • a bottleneck for a large number of peers (performance degradation) • lacks scalability and robustness • examples: • scientific computation - SETI@home, BOINC, Folding@home, Genome@home • digital content sharing - Napster, Openext • others - Jabber (IM), Net-Z and StarCraft (entertainment), etc. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 39/116 Taxonomy of P2P Systems Centralized P2P Systems Taxonomy of P2P Systems III Centralized P2P Systems II. Peer D Central Server Peer C — —> Peer A Peer E 9- Peer B f / Figure: Centralized P2P Systems: Peer A submits a request to the central server to acquire a list of nodes that satisfy the request. Once it obtains the list (which contains Peers B and C), it communicates directly with them. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 40/116 Taxonomy of P2P Systems Decentralized P2P Systems Taxonomy of P2P Systems III. Decentralized P2P Systems I. Decentralized (Pure) P2P Systems • peers have equal rights and responsibilities • each peer has only a partial view of the P2P network and offers data/services that may be relevant to only some queries/peers • ^> locating peers offering services/data quickly is a critical and challenging issue • advantages: • immune to single point of failure • (usually) provide high performance, scalability, robustness, and other desirable features • examples: Gnutella, Crescendo, PAST, FreeNet, Canon, etc. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 41/116 Taxonomy of P2P Systems Decentralized P2P Systems Taxonomy of P2P Systems III Decentralized P2P Systems II. Two dimensions in the design of decentralized P2P systems: • flat (single-tier) vs. hierarchical (multi-tier) network structure • flat structure —>* the functionality and load are uniformly distributed among the participating nodes • hierarchical structure —>* multiple layers of routing structures • example: national level (interconnecting states), states level (interconnecting universities), universities level (interconnecting departments), etc. • offers certain advantages (fault isolation and security, effective caching and bandwidth utilization, hierarchical storage, etc.) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 42/116 Taxonomy of P2P Systems Decentralized P2P Systems Taxonomy of P2P Systems III. Decentralized P2P Systems II. • structured vs. unstructured logical topology • unstructured P2P system —>• each peer is responsible for its own data, and keeps track of a set of neighbors that it may forward queries to • no strict mapping between the identifiers of objects and those of peers • =4> locating data is a challenge (its difficult to precisely predict which peers maintain the queried data) • =^> there is no guarantee on the completeness of answers (unless the entire network is searched) • =4> there is no guarantee on response time (except for the worst case where the entire network is searched) • structured P2P system —>> data placement is under the control of certain predefined strategies (generally, a distributed hash table - DHT) • there is a mapping between data and peers • =4> these systems can provide a guarantee (precise or probabilistic) on search cost • =^> however, typically at the expense of maintaining certain additional information • (systems employing a mix between structured and unstructured topology also exist) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 43/116 Taxonomy of P2P Systems III. Decentralized P2P Systems III. Peer D Peer H Figure: Decentralized P2P Systems: Peer A requests for some data that Peer D and Peer H have. The query will be broadcasted to the neighbors of Peer A, and gradually, to the other peers in the whole network (Gnutella). Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 44/116 Taxonomy of P2P Systems Hybrid P2P Systems Taxonomy of P2P Systems III. Hybrid P2P Systems Hybrid P2P Systems 9 the main advantage of centralized P2P systems: quick and reliable resource locating • BUT with the limitation in terms of scalability * the main advantage of decentralized P2P systems: scalability • BUT with the limitation in terms of longer time necessary for resource locating • =4> Hybrid P2P systems: 9 to maintain the scalability, there are no central servers • however, more powerful peer nodes are selected to act as servers to serve others • = super peers • resource locating can be done by both decentralized and centralized search techniques (asking super peers) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 45/116 Taxonomy of P2P Systems III. Hybrid P2P Systems III. Figure: Hybrid P2P Systems: At first, the query is forwarded to a superpeer/ultrapper node, which the query node belongs to; the superpeer in cooperation with other superpeers looks for the superpeer (lookup based on a structured algorithm), which maintains a node having an answer for the request. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 46/116 Lecture Overview Q Client-Server vs. Peer-to-Peer • Client-Server Systems • P2P Systems • Comparison 0 Generic P2P Architecture • Overlays and Peer Discovery • Service/Resource Discovery Q Taxonomy of P2P Systems • Centralized P2P Systems • Decentralized P2P Systems • Hybrid P2P Systems Q Routing in P2P Networks • Introduction, Motivation • Routing in Unstructured P2P Networks • Routing in Structured P2P Networks • Routing in Hybrid P2P Networks & Information Sources Routing in P2P Networks > routing of messages/requests is one of the key operations in P2P systems • to locate desired resources, each peer should be able to forward queries to a subset of neighbor peers that are closer to the destination than any other peer • —>* the design of routing protocols is one of the most widely researched issues > the key differences between the various schemes lie in the amount of information (metadata) being maintained at each peer • and how this information is organized • no metadata there is no other way for locating information except for flooding/broadcasting the request through the network Perfect information No information (Napster-like) (Gnutella-like) Structured Unstructured Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 48/116 Routing in P2P Networks The Lookup Problem Put (Key="title" Value=file data...) Publisher Client Get(key="title") Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 49/116 Routing in P2P Networks The Lookup Problem - Centralized Lookup (Napster) SetLocftitle", N4) Publisher@|\L Key="title" Value=file data... DB N Client Lookupftitle") N9 N7 Ns N. 7 Simple, but O(n) state information has to be maintained on a single central node, and the network suffers from a single point of failure. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 50/116 Routing in P2P Networks The Lookup Problem - Flooded Queries (Gnutella) N Publisher@N4 Key="title" Value=file data... N N Lookupftitle") Client Robust, but in the worst case O(n) messages has to be transmitted per lookup. Autumn 2018 51/116 Eva Hladká (Fl MU) 5. P2P networks Routing in P2P Networks The Lookup Problem - Routed DHT Queries (Chord, CAN, Pastry, Tapestry, .. .) Publisher Key=H(audio data) Value={artist, album title, track title} N Client Lookup(H(audio data)) N 8 Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 52/116 Introduction, Motivation Routing in P2P Networks Evaluation Metrics The effectiveness/efficiency of a routing scheme can be evaluated by several metrics: • Storage • each peer may need to incur some storage space for maintaining metadata (used for searching) • storing more metadata it is more costly to keep these data up-to-date • Efficiency • a system is efficient if it can locate the resources quickly • metric of efficiency is the response time (can be measured by the average query • Usability • reflects the ease of use, and the types of queries that can be supported • e.g., depending on the metadata maintained, one system may support complex • Coverage • refers to whether the search space contains the answers • a scheme with a higher coverage is certainly more useful • Scalability • important - makes the routing scheme useful in largescale environments • a measure of scalability - e.g, the number of messages that need to be routed in path length) queries, while another one can perform an exact match only order to locate information Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 53/116 Routing in P2P Networks Routing in Unstructured P2P Networks uting in Unstructured P2P Networks each peer typically stores its own data objects and selfmaintains a set of links to neighbor nodes when a node wants to join the system, it simply contacts an existing node and copies links of that node to form its own links • (later maintained independently on the contacted node) =4> no peers have global knowledge of data placement • flooding-based techniques have to be used for queries • to alleviate the problem of flooding the system with query messages, a Time-to-Live (TTL) value is usually attached to each query • the challenge is, how to optimize query processing in the limited number of search steps constrained by TTL several routing strategies have been proposed: • Breadth-First Search (BFS) - e.g., Gnutella • Depth-First Search (DFS) - e.g., FreeNet • Heuristic-Based Routing Strategies Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 54/116 Routing in P2P Networks Routing in Unstructured P2P Networks Depth-First Search (DFS) Algorithm 1 : FreeNet_Search (Node x, Key k, TTL t) 2 3 4 5 6 7 8 9 10 11 12 13 14 result = Local_Search(k) if result —found then return result to the requester node else if t = 0 then return "not found" to the requester node else repeat pick a neighbor node y in the routing table of x that has the nearest key to k and has not been searched before result = FreeNet_Search(>% k, t — 1) until result = found or all neighbors have been searched return result to the requester node end if end if Figure: FreeNet's routing strategy: instead of sending a query to all neighbors, each node selects the most promising neighbor that can answer the query and sends the query to only that node. If the node does not receive a reply within a certain period of time (or the answer cannot be found), the node selects a next promising neighbor. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 55/116 Routing in P2P Networks Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Iterative Deepening • the idea: • a query is initiated with a sequence of multiple traditional BFS searches by enlarging search radius gradually • the search process terminates when either the maximum depth is reached or the results for the query satisfy user's requirements 9 algorithm details: • a system policy P must be provided to specify the sequence of the depths at which the iteration happens • P = Di, D2,..., Dn, where D\ < D2 < ... < Dn • under this policy, the source node first sends a query message to the network via BFS search of depth D\ • if the result obtained satisfies user's requirements, the query is terminated • otherwise, the source node issues another resend query message (with the same query ID) with a BFS depth of D2 • the nodes that are less than Di-hops away from the source node do nothing but just forward the query to their neighbors • the further nodes process the query in the same way as in the first iteration • similarly for D3, D4, etc. • if the query is not answered until the depth of Dn, the search process terminates Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 56/116 Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Directed BFS and Intelligent Search I. • the idea: • in BFS, each node sends the query to all of its neighbors • in Directed BFS, each node only queries a subset of its neighbors • the key point is how to intelligently choose "good" neighbors that would potentially contribute more relevant results for the query • details on choosing the neighbors: • each node maintains some statistics of its neighbors: the number of previously answered queries through a neighbor node, the number of results obtained, and the latency in receiving the results • based on these statistics, the node can choose the neighbors "intelligently" based on several heuristics, e.g.: • choose the one that returned the largest number of results previously • choose the one that incurred the least hop-count messages previously • choose the one that forwarded the largest number of messages previously • choose the one that have shortest message queues • etc. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 57/116 Routing in P2P Networks Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Directed BFS and Intelligent Search II. Directed BFS • advantage: the number of query messages in the network is greatly reduced as compared to standard BFS technique • disadvantage: the statistics stored about each neighbor are too simple • they do not contain information related to the content of queries • =4> Intelligent Search • each peer ranks its neighbors based on their relevances to the query • the query is routed only to those neighbors that have high relevances • it thus provides a more exact ranking of peers than Directed BFS • has good performance in networks that exhibit a high degree of query locality Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 58/ Routing in P2P Networks Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Local Indices Search • the idea: • each node creates and maintains indices for both its local data and the data on its neighbor nodes that are within a radius of k hops from it 9 if k = 0, this method is similar to BFS search (local data index only) • the result returned at such a node is the same as the result, which would be returned by processing the query at all the nodes within a radius of k hops from the node 9 details: • the queries are processed based on a global policy P that specifies a list of depths in the search tree where the query is processed • just the nodes located at the depth specified in P process the query o the other nodes simply forward the query to their neighbors (without processing it) 9 advantage: • reducing the processing cost by limiting the query processing to fewer nodes • disadvantages: • higher storage cost (more indices need to be stored at a node) • higher update cost for these indices • inconsistency/obsolescence of the indices (due to dynamics of the network) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 59/116 Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Random Walk I. • the idea: • when a peer issues/receives a query, it randomly selects a neighbor to send or forward the query to • this process repeats until the search result is found • the main disadvantage: • it suffers from long delays in query processing • k-walker Random Walk Algorithm • the query initiator (the source node) sends k query messages to its randomly-selected neighbors (instead of just a single one = the original 1-walker algorithm) • when a node receives a query message (a walker), it just follows the basic random walk to randomly select a single neighbor to forward the query to 9 the number of messages (visited nodes) increases linearly as compared to the 1-walker algorithm Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 60/116 • or TTL expires (if employed) the result is not found • details: Routing in P2P Networks Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Random Walk II. • details cont'd.: • Random Breadth First Search (RBFS) • similar to the /c-walker Random Walk o the query initiator first randomly selects a subset of its neighbors to send the query to • each of these neighbors then randomly selects a subset of its neighbors, where the query is forwarded • etc. • the number of messages (visited nodes) increases exponentially as compared to the 1-walker algorithm Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 61/116 Routing in P2P Networks Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Adaptive Probabilistic Search (APS) • the idea: • a search method that combines techniques of both /c-walker random search and probabilistic search • the main difference between APS and random walkers: 9 random walkers send the query to random neighbors while APS sends the query to neighbor nodes based on some probabilities • each peer contains a probability for each neighbor with respect to each object (determined from past results) • details: • two approaches to update the probabilities: • Optimistic approach - the system proactively increases the probabilities for selected (= queried) neighbors along the search path and decreases their probabilities only if the walker passing through them terminates with a failure • Pessimistic approach - the system proactively decreases the probabilities for selected (= queried) neighbors along the search path and increases their probabilities when the walker passing through them terminates with a success • swapping-APS - each peer swaps between optimistic and pessimistic method • based on an observation of the ratio of successful walkers for each object • weighted-APS - takes into account the location of objects Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 62/116 Routing in P2P Networks Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Interest-Based Shortcuts I. • the idea: • each peer adds additional links on top of an existing searching network to improve the search performance 9 these links (called interest-based short-cuts) connect two peers having a similar interest • details: 9 when a peer issues a query, it first employs interest-based shortcuts to forward and process the query • if the result is found, the search terminates 9 otherwise, the normal query processing algorithm is used • shortcut construction: 9 when a peer joins the system, it has no shortcuts 9 after each successfully processed query, the query initiator adds shortcuts to peers providing the answers for that query 9 each peer stores only a limited number of shortcuts that have the highest utility (due to space constraints) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 63/116 Routing in P2P Networks Routing in Unstructured P2P Networks Heuristic-Based Routing Strategies Interest-Based Shortcuts II. Routing in P2P Networks Routing in Structured P2P Networks Routing in Structured P2P Networks • the unstructured P2P networks suffer from the problem of low searching efficiency • unlike in the unstructured P2P systems, participant nodes in a structured P2P system are required to organize into some fixed topologies • such as a ring (Chord), a multidimensional grid (CAN), a mesh (Pastry and Tapestry), or a multiple list (Skip Graph) • =>> when a node joins the system, it has to follow some strict procedures to set up its position • can be guaranteed, that if a result of a query exists in the system, it will be found • moreover, in an efficient way - most systems can provide an answer for a query within 0(log N) steps/messages (A/ = number of nodes) • disadvantage: • the need for a network topology incurs high maintenance cost (changes in routing tables) • based on the overlay network structure, structured P2P systems can be classified into the following categories: • Distributed Hash Table (DHT) based systems - e.g., Chord, CAN, Tapestry and Pastry, Viceroy and Crescendo, etc. • Skip List based systems - e.g., Skip Graph, SkipNet, etc. • Tree based systems - e.g., P-Grid, P-Tree, BATON, etc. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 65/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Distributed Hash Table every node in the P2P network manages its part of global hash table storage/retrieval of an item s means quering the node, which manages the part, where the hash(s) belongs to Data A Data B Data C Data D Data E DHT Klíč: 0x45A23 Klíč: 0x8C39A Klíč: 0xBF4D2 Klíč: 0x6C561 Klíč: 0x6C563 Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 66/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Chord I. • one of the most widely known routing mechanism in structured P2P networks • the idea: • uses a one-way consistent hash function to map each node and data item to an /77-bit identifier in a single-dimensional identifier space • the hash function uses the node's IP address to generate an identifier for a node, and • the data item (or the key of the data item) to generate an identifier for the data • the identifier space must be chosen large enough (the probability of assigning the same identifier to different nodes should be negligible) • details: • the identifier space is a circle of numbers from 0 to 2m — 1 • the system assigns a key k to the first node n whose identifier is equal to or follows the identifier k in the circle space • i.e., the key k is assigned to the first node clockwise from k Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 67/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Chord II. IP="202.120.224.102 Key="Sailing Figure: An identifier circle based on consistent hashing - keys Ke and Kis are assigned to the same node identifier A/30 (obtained by hashing the IP address "202.120.224.102"). The key Kse (obtained by hashing the word "Sailing") is assigned to the node identifier A/70; the key /C100 is assigned to the node identifier A/115; the nodes A/42 and A/120 store no data items. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 68/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Chord III. - Simple lookup algorithm Simple lookup algorithm: • each node only needs to know its immediate successor node • when a node receives a query request: • first, it checks its local storage to see if it holds the queried data item • if yes, the result is returned to query sender • if no, it forwards the query to its immediate successor node • the lookup terminates, when • the result is found • the identifier of a node's immediate successor exceeds the identifie the queried data item ^> the result cannot be found • the complexity is O(N) (N = the number of nodes in the system) N N30 Where is Key 56? K, Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Chord III. - Scalable lookup algorithm I. Scalable lookup algorithm: • instead of maintaining only a single immediate successor node, each node maintains a finger table consisting of m successor nodes • when a node n receives a query request: • if the node does not hold the queried data, it searches its finger table for a node r! with the highest node identifier that satisfies the condition n.id < n'.id < k • if such a node exists, the node n asks n' to find the key k • otherwise, the node n asks its immediate successor to find k • the lookup terminates, when • the result is found 9 the identifier of a node's immediate successor exceeds the identifier of the queried data item ^> the result cannot be found • the complexity is 0(log N) (N = the number of nodes in the system) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 70/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Chord III. - Scalable lookup algorithm II. M 9NiN. 111120 N * ■70 n 63 Finger Table N 30 N 3„ N 30 N7-s N 30 N 30 ^7 + 32 M7+64 Where is Key 117? ■70 m 63 Figure: An example of finger table entries (left) and an example of a routing path for key Kuj starting at node A/7 (right). Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 71/116 Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Chord IV. System construction: • when a new node joins the system, it needs to: O find its position in the Chord ring and obtain data it should be responsible for (based on keys) Q initialize its finger table O update finger tables of other nodes to reflect the presence of it • when an existing node leaves the system, it does not need to do anything Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 72/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Content Addressable Network (CAN) I. • the idea: • a routing system built on a virtual d-dimensional Cartesian coordinate space o the system partitions the storage space into different zones, each of which is assigned to a node • such a node stores all data items belonging to its zone • the system uses a uniform hash function to map the data key value to a point p in the coordinate space (thus obtaining a d-tuple) • details: • inserting a data item: Q the data key value is mapped into a point p in the coordinate space O the node n, whose zone covers p, is found and contacted to store the new data item • processing a query is similar • if the result exists, it should be stored on the node covering the particular zone • each node needs to maintain information about its neighbor nodes • i.e., the nodes covering adjacent zones • the routing is based on a simple greedy forwarding algorithm _9 in every step, a node having closer coordinates to the destination zone is chosen_ Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 73/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Content Addressable Network (CAN) II. (0.5-0.75,0.5-1.0) c (0.0-0.5,0.5-1.0) D E — A (0.0-0.5,0.0-0.5) B (0.5-1.0,0.0-0.5) >(0.75-1.0,0.5-1.0) A Node E's Virtual cordinate zone E's neighbors: D and B Figure: A CAN system using two dimensional space with 5 nodes. Eva Hladká (Fl MU) 5. P2P networks Autumn Distributed Hash Table (DHT) based P2P systems Content Addressable Network (CAN) III. + + * + + ^ A + + A 1 1 + + + ^ _A_ 0 0,5 More paths might be used to reach the destination Looking for the data item having the key (0,6;0,8) Figure: An example of a data item lookup in a CAN system. Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Content Addressable Network (CAN) IV. System construction: • when a new node joins the system, it needs to: O find an arbitrary node, which is already connected to the network O identify a zone, which might be divided, and ask its owner/maintainer node to split the zone into two parts • the original node keeps maintaining one part, the new node starts to maintain the second one O construct its own routing table and update the routing tables of its neighbors • when an existing node leaves the system, it has to ask its neighbor to merge the zones into a single one Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 76/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Pastry I. • the idea: • a routing system based on PRR trees o PRR = Plaxton, Rajaraman, and Richa (1997) • a node identifier is an m-bit number broken up into a sequence of digits having the base 2b • e.g., a 128-bit identifier is broken up into 32 4-bit digits (b = 4, base = 24 hexadecimal sequence of digits) 9 b . . . configuration parameter • a data item is stored on a node having the identifier, which shares the longest prefix with the data identifier 9 in every routing step, a neighbor node having a longer prefix in common with the destination node (longer by 1 digit, i.e., b bits) is chosen 9 the routing complexity is O(log2b N) • details: • each peer has a routing table to route messages 9 organized in a fixed number of levels (= |7og"2b(A/)]) and within each level a fixed number of entries (= 2b — 1) • row ID = the length of prefix in common with the destination node • column ID = next possible step Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 77/116 Routing in P2P Networks Routing in Structured P2P Networks Distributed Hash Table (DHT) based P2P systems Pastry II. 0 1 2 ;< 4 S / 8 9 a b f d e t X X v .V X \ \ x X X x X x - — — — _ TT ~6~ 6 6 6 6 6 6 ti 6 6 ŕ, 6 6 ti 0 1 2 3 4 6 7 8 a b r d v r \ v \ .\ x x \ x \ x v x A — — - 6 6 1i 6 6 h 6 6 6 (r h tí 6 u S :> $ J $ 5 S S s 5 S s S S S \s 0 í 2 3 4 m H 7 8 9 b c d f r X X X X X X x x x x x x x x x ._ ^—— - _ _ 6 6 6 6 6 6 6 6 6 6 6 6 ti 6 6 5 5 h b h 5 ,> r a a a Ú a a a a a a a O(log n) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 85/116 Skip List based P2P systems Skip List structure III. head Search for key *R' —* success tail ,_, „_„ ......► failure .—. Level 0 Level 1 Level 2 J A A J \ M J M -► ' R ' W -00 +00 Figure: The searching process in a Skip List structure. Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems Skip Graph I. the idea: • a routing system based on Skip Lists • pure Skip Lists are not suitable, since the top-level nodes may become overloaded • unlike pure Skip List, which has only one list at each level, a Skip Graph has many lists at each level • each node participates in a list at each level • the system controls the lists, which a node belongs to, by a random membership vector (created when the node joins the system) • the number of levels is O(log N) lookup details: • once a node issues a query: • the search process always starts at the highest level of that node • at each step, if there is a neighbor node at the same level that keeps a closer value to the search key, the node forwards the query to that neighbor • otherwise, the node continues the search process at a lower lever • the destination node containing the result is found when the search process reaches the bottom level • the query processing complexity is 0(log N) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 87/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems Skip Graph II. CM > A J 100 101 000 001 Oil 110 > AA A 100 J 10 W 01 > 001 001 100 001 Oil \Membership vectors 001 on 110 101 The Membership vector only defines, which lists the particular element belongs to (the lists are sorted by a data key). Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 88/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems Skip Graph Restricting to the lists containing the starting element of the search, we get a skip list (the pure skip list searching method can be used then) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 89/ Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems Skip Graph IV. System construction: • when a new node (having an identifier X) joins the network: • based on its membership vector m(X), X joins the lists of nodes whose membership vector shares the same prefix with m(X) at different lengths • in particular: 9 X first joins the list at level 0 (to the nodes containing keys closest to the X's key) • for every level / > 1, X links to the closest node Y having the same /-length prefix with the node X Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 90/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems Skip Graph V. CM 1 001 .........■■■ ■ ■ 001 100 100 ■ ■ ■ ■ L L I I J ■■■■■■■■■■ L I I J J ■■■■■■■■■■ I I I J ■ Oil Oil 110 c ■....................|l i 110 101 4-» 110 101 W 101 new node 001 Figure: Step 1: Starting at an arbitrary node, find a nearest (data) key at level 0. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 91/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems Skip Graph VI. CM > & W r -- M ■ A 100 J 101 -■ 001 001 Oil 110 > 100 M 110 01 O; "5: 01 01 011 1 k -J ■ M w r r & J R F -■■ ■■ ■■ 100 001 011 110 101 001 Figure: Step 2: At each level /, connect to the list with a matching prefix of membership vector of length /. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems SkipNet I. the idea: • a routing system very similar to Skip Graph • instead of Skip Lists, the SkipNet organizes nodes into rings 9 similarly to the Skip Graph, organized into levels as well • the nodes are sorted on each level based on a data key 9 on a particular level, every node has a pointer to its neighbors stored in its routing table • the pointers on the level h point to the nodes that are roughly 2h nodes to the left and right of the given node • all the nodes are connected by the root ring formed at level 0 • the routing/lookup mechanism and system construction are very similar to the Skip Graph's ones Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 93/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems SkipNet II. Ring 000 Ring 001 Ring Ring 010 011 Ring 100 Ring 101 Ring 110 Ring 111 Ring 10 Ring 11 L = 3 L = 2 Ring 1 M O V L = 1 Root Ring Level: L = 0 X V Figure: The full SkipNet routing infrastructure for an 8 node system, including the ring labels. Autumn 2018 94/116 Eva Hladká (Fl MU) 5. P2P networks Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems SkipNet III. M D A \ v 1 / / / / jl ^ Z . _.__v T s JL Level o G Level o G 2 T T 2 D D 1 M X 1 Z O 0 D Z 0 X T Figure: The routing tables for nodes A and V. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 95/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems SkipNet IV. A routing example: Routing from A to V Ring Ring Ring Ring Ring Ring Ring Ring 000 001 010 011 100 101 110 111 OO(!)QDQD0 - L =2 L =1 L=0 Figure: At first, the message is forwarded to a neighbor closer to the destination Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 96/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems SkipNet V. A routing example: Routing from A to V Ring Ring Ring Ring Ring Ring Ring Ring 000 001 010 011 100 101 110 111 ©O(5)©OQO0 ^ L =2 L =1 L=0 Figure: Node T's routing table. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 97/116 Routing in P2P Networks Routing in Structured P2P Networks Skip List based P2P systems SkipNet VI. A routing example: Routing from A to V Rine Rlná RlnĚ Rir1g RlnĚ Rin^ Rlng RiľlÉ 00^ 001 01Ů Oil 100 101 110 111 Level: L = O Figure: Since there is a direct access to the node V at level 0, the lookup terminates. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 98/116 Routing in P2P Networks Routing in Structured P2P Networks Tree based systems P-Grid I. the idea: a the P-Grid is based on a virtual binary tree structure in which each peer maintains a leaf node of the tree • the system assigns each peer an identifier, which is the binary bit string representing the path from the root to the leaf node • each peer is then responsible for all data items whose prefix is equal to the peer identifier • for fault-tolerance purposes, multiple peers can be assigned the same identifier • for routing purposes, each peer further maintains a routing table Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 99/116 Routing in P2P Networks Routing in Structured P2P Networks Tree based systems P-Grid II. Data with] [Data with ^prefix 00J ^prefix 00 Data with ^prefix 01^ Data with| [Data with ^prefix 10J ^prefix 10^ [Data withr ^prefix 11 j A peer node Routing table (route keys with prefix P to peer X) Data store (keys have prefix P) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 Routing in P2P Networks Routing in Structured P2P Networks Tree based systems P-Grid III. the routing/lookup mechanism: • when a peer n receives a query having the key k, it checks, whether its identifier is a prefix of k • if yes, it searches its local storage to find the result • if no, the peer looks up its routing table to find a closer neighbor node to forward the query • the maximum number of search steps is bounded by the height of the tree the lookup performance is O(log2 N) Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 101 / Routing in P2P Networks Routing in Structured P2P Networks Tree based systems P-Tree I. the idea: • in P-Grid, the balance of the tree structure cannot be guaranteed • P-Tree is based on a virtual balanced B+-Tree built on top of a Chord ring a each peer maintains: a Chord node, which is a leaf node of the tree structure, and a semi-independent B+-Tree, which is a peer's view of a fully independent B+-Tree • a fully independent B^-Tree at a peer is a £+-Tree, where the value stored at the peer is considered as the smallest value in the Chord ring • a semi-independent B^-Tree contains all nodes in the leftmost root-to-leaf path of the corresponding fully independent 8+-Tree • to make it easy for maintenance, ranges of B+-Tree nodes can be overlapped (see node C in the following figure) 9 9 Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 102/116 Routing in P2P Networks Routing in Structured P2P Networks 42 7 23 30 B D F 42 30 42 7 H B 30 31 G 5 29 31 E G 5 7 13 23 B C D 31 7 29 B E 31 42 5 H A 7 29 31 5 E G A 7 13 23 29 C D E 13 29 42 7 E H B 13 23 29 30 D E F 29 42 7 H B 29 30 31 F G 23 31 5 G A 23 29 30 E F (a) Semi-independent ß+-Trees maintained at P-Tree nodes. Eva Hladká (Fl MU) 5. P2P networks 5 7 13 23 B C D 5 29 31 "' E G % 29 30 E F (b) The fully-independent B at node A. Autumn 2018 Routing in P2P Networks Routing in Structured P2P Networks Tree based systems BATON I. the idea: o in comparison with standard tree-based structures, the BATON provides two main features: • data is stored at both leaf nodes and internal nodes • in addition to parent and child links, nodes in the BATON network also have adjacent links and neighbor links • adjacent link is used to connect a node to a node maintaining an adjacent range of values (adjacent to the range the node maintains) • neighbor link is used to connect a node with its neighbors (at the same level in the tree structure) having a distance 2', / > 0 from the node • the purpose of these links is to avoid the bottleneck problem at the root of the tree structure in query processing Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 104 /116 Adjacent link O Sample node O Parent node Qchild node O Neighbor node Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 105/116 Routing in P2P Networks Routing in Structured P2P Networks Tree based systems BATON III. lookup details: • when a peer x receives a query: O if the searched key falls into the range of values managed by x, it responds to the query O otherwise, it forwards the query to the farthest neighbor that is nearer to but not overshooting the searched key O if such a neighbor does not exist, x forwards the query to either a child (if it exists) or an adjacent node of x in the search direction Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 106 /116 [0-5) [8-12) [17-23) [38-41) [50-54)[57-61)[64-67)[69-73) [79-83) [86-90)[95-100) Figure: A lookup example in BATON: the node H wants to search for a data item (having the key 74) stored in the node C. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 107/116 Routing in P2P Networks Routing in Hybrid P2P Networks Routing in Hybrid P2P Networks • hybrid P2P systems organize the peers into a hierarchical network o powerful peers (superpeers, supernodes) lie in a high level, and • common peers (also named client peers) lie in lower levels • each common peer belongs to a supernode and does not connect with any other common peer that does not belong to the same supernode <* the general routing scheme in hybrid P2P networks: O a client peer sends a query to its supernode 0 the supernode searches its directory to determine which client peer or supernode has the desired answers O the query is sent to the supernode that may have the desired answers • it uses its directory of all its client peers to answer the query O the IP address of the client peer having the desired answers is returned to the query peer • the query peer exchanges resources with that peer • examples: • KaZaA, BestPeer, Edutella, etc. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 108/116 Routing in P2P Networks Routing in Hybrid P2P Networks Routing in Hybrid P2P Networks Edutella Figure: The Edutella network structure. A query routing in Edutella is first directed to superpeers in HyperCuP network (= HyperCube P2P network), where the suffix-based routing scheme could be employed. Autumn 2018 109/116 Eva Hladká (Fl MU) 5. P2P networks Routing in P2P Networks Routing in Hybrid P2P Networks Routing in Hybrid P2P Networks Ultrapeers Figure: The modified Gnutella network with ultrapeers. Suppose that the resources requested by peer C12 are located at the peer C9: the peer C12 first requests its ultrapeer 1/4, then 1/4 floods the query to U2 via ill; ill searches its reflector index and finds that C9 has the desired answers —>> it sends the IP address of C9 back to C12. Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 110/116 Routing in P2P Networks Routing in Hybrid P2P Networks Routing in Hybrid P2P Networks Structured Superpeers Figure: The structured superpeers: the superpeers SO, SI, S2, and S3 control ranges (0,4], (4,8], (8,12], and (12,0], respectively. If the peer PI requests key = 10, it first sends the lookup key to SO; SO relays the key to S2 (since S2 controls the range where the key belongs), which replies the query initiator with the IP of the relevant node gtnrincr thp rpmipctprl rlata 111/116 Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 Routing in P2P Networks P2P Routing Conclusion P2P Routing Conclusion Structured vs. Unstructured P2P Networks Comparison structured P2P unstructured P2P routing based on a routing table flooding, random walk, ... lookup possibilities based on keys only possibility to ask more compex queries existing item is always found yes cannot be guaranteed critical part node join/disconnect lookup/routing Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 112/116 P2P Routing Overview 1. Conclusion System Overlay network Routing table Routing method Gnutella Unstructured, Random neighbors Breadth First Search with Random Time-to-Live topology FreeNet Unstructured, Random neighbors Depth First Search with Random Time-to-Live topology Chord Structured, Ring Neighbors at Repeatedly jump to the topology distances 2' in the farthest node in the routing ring table whose id is still less than the search key CAN Structured, Mesh Neighbors at Repeatedly travel through topology adjacent positions in the neighbor that is closer to the mesh the destination Pastry & Tapestry Structured, PRR Neighbors sharing Repeatedly forward the tree topology common prefix message to the neighbor identifier at different having the longest matching levels prefix identifier Viceroy Structured, Five neighbors: one Three steps: going up, going butterfly at the upper level, down, and vicinity search topology two at the lower level, and two at the same level Crescendo Structured, Chord-like A combination of Chord-like hierarchical ring neighbors at routing and the routing topology different ring levels between rings at different levels Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 113/116 Routing in P2P Networks P2P Routing Conclusion P2P Routing Conclusion Overview II. System Overlay network Routing table Routing method Skip Graph Structured, Neighbors sharing Travel from the highest to multiple linked common prefix the lowest level of the list. At lists topology membership vector each level, jump to the at different lengths neighbor closer to the destination if such a neighbor exists SkipNet Structured, Neighbors are Skip Graph-like routing, hierarchical ring predecessors and traveling from the highest to topology successors at the lowest level of the ring. different ring levels P-Grid Structured, A neighbor at the Travel from the root to the binary tree other side of the tree leaf. At each level, jump to topology rooted at each the neighbor closer to the internal node from destination the root to the leaf P-Tree Structured, a Neighbors are nodes Travel from the root to the combination of a in the left-most leaf. At each level, jump to B+-Tree and a root-to-leaf path of the neighbor closer to the Chord ring the B+-Tree destination topology BATON Structured, Neighbors are If not having full routing balanced tree parent, children and tables, go to parent. topology Chord-like Otherwise, go to the neighbors at the neighbor or the child closer same level to the destination Edutella & Ultrapeers Hybrid, a Neighbors exist A client peer always routes combination of only at superpeer its requests to its superpeer structured and level. At client side, while routing at supper peer unstructured each client peer level depends on the topology connects to a topology employed at that superpeer level Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 114/116 Information Sources Lecture Overview I Q Client-Server vs. Peer-to-Peer • Client-Server Systems • P2P Systems • Comparison 0 Generic P2P Architecture • Overlays and Peer Discovery • Service/Resource Discovery Q Taxonomy of P2P Systems • Centralized P2P Systems • Decentralized P2P Systems • Hybrid P2P Systems Q Routing in P2P Networks • Introduction, Motivation Routing Routing Routing n Unstructured P2P Networks n Structured P2P Networks n Hybrid P2P Networks Q Information Sources Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 115/116 P2P Information Sources Fl courses: a PA128: Similarity Searching in Multimedia Data (prof. Zezula) Literature: • 0. H. Vu at al. Peer-to-Peer Computing: Principles and Applications. Springer, 2010 • Milojicic et al. Peer-to-Peer Computing. HP Labs, 2002 9 D. C. Verma. Legitimate Applications of P2P Networks. Wiley, 2004 • X. Shen, H. Yu, J. Buford, M. Akon. Handbook of Peer-to-Peer Networking. Spriger, 2010 • J. Buford, H. Yu, E. K. Lua. P2P Networking and Applications. Morgan Kaufmann, 2009 Eva Hladká (Fl MU) 5. P2P networks Autumn 2018 116/116