p2p: systems and applications Internet Avanzado, QoS, Multimedia Carmen Guerrero

Transcription

1 p2p: systems and applications Internet Avanzado, QoS, Multimedia Carmen Guerrero Dpto. Ingeniería Telemática Index Introduction Taxonomy Classification of p2p overlay networks Applications Voip in p2p -> p2psip Estandardization Bibliography Work proposals 2 1

2 Introduction P2p overlay nets are: Distributed systems in nature Without any hierarchical organization Without any centralized control Features: Robust wide-area rounting architecture Efficient search of data items Selection of nearby peers Redundant storage Hierarchical naming Trust and authentication Anonymity Massive scalability Fault tolerant Self organization 3 Abstract p2p overlay network architecture Fuente: [Survey05] 4 2

3 First classification of p2p overlay nets Structured Topology tightly controlled Content is not located randomly but at specific peers That means more efficient queries How? DHT (Distributed Hash Tables) Data object location information is placed deterministically at the peers with IDs corresponding to the data object s unique key Assigned uniform random peerid Large space of identifiers Data objects are assigned unique identifiers called keys Keys are mapped to a unique live peer in the overlay net Scalable storage and retrieval of {key,value} Each peer maintains a small routing table consisting of its neighboring peer Differente DHTs: data organization, key space and routing strategies. Efficiently locate rare items More overhead than unstructured p2p for popular content 5 First classification of p2p overlay nets Unstructured Napster (1999).- p2p file sharing and a centralized file search facility Self scaling: more peers, more aggregate download capability It does not requiere much bandwidth for the central server Single point of failure Gnutella (2001). Decentralized system. Both search and download capabilities First example of unstructure p2p No topology Flooding messages When a peer receives a query, it responds with the list of content that maches the query Effective for locating high replicated items Resilent to peer joining/leaving the peer the system Not suited for locating rare items, better for popular content Not scalable: load of each peer grows linearly with the queries and nº of peers Peer readily become overloaded Today, decentralized unstructured p2p network are commonly used over Internet. Recent effort on KBR (Key Based Routing) versus ad-hoc nature in unstructure p2p. 6 3

4 Taxonomy Decentralization Architecture Lookup protocol Systems parameters Routing performance Routing state Peer join and leave Security Reliability and fault resiliency 7 Structured CAN (2000) Chord (2003) Tapestry (2004) Pastry (2001) Kademlia (2002) Viceroy (2002) 8 4

5 CAN Content Addressable Network Hash-table functionality to an Internet-like scale Scalable, fault-tolerant and self-organizing Architectural design: Virtual d-dimensional cartesian coordinate space The entire coordinate space is dynamically partitioned among all the peers (N peers) Every peer possesses its individual, distinct zone within the overall space A CAN peer maintains a routing table: IP address and virtual coordinate zone of each of its neighbours in the coordinate space. A peer routes a message toward destinatipn forwarding to the neighbor peer that is closest to the destination coordinates Routing performance ofθ(d x N 1/d ) Routing state 2 x d bound 9 CAN Fuente: [Survey05] 10 5

6 CAN The virtual coordinate 2-d space is used to store {key K, value V} Key K is deterministically mapped onto a point P in the coordinate space using a uniform hash function The lookup protocol retrieves an entry corresponding to key K, and any peer can apply the same deterministic hash functionto map K onto a point P and then retrieve the corresponding value V from the point P. If the requesting peer or its inmediate neighbour to not own the poing P, the request must be routed through the CAN infraestructure until it reaches the peer where P lays. A peer maintains the IP addresses of those peers that hold coordinate zones adjoining its zone. A new peer that joins the systems must have its own portion of the coordinate space allocated. By splitting an existing peer zone in half 11 CAN CAN has an associated DNS domain name that is resolved into the IP address of one or more CAN bootstrap peers (which maintain a partial list of CAN peers) A new peer to join, it looks up in the DNS of a CAN domain to retrieve a bootstrap peer s IP address The bootstrap peer supplies the IP address of some randomly choosen peer in the system The new peer randomly chooses a point P and send a JOIN request destinated for the point P Each CAN peer uses the routing mechanism to forward the message until it reaches the peer in which zone P lies. The current peer in zone P then split its zone in half and assigns the other half to the new peer. 12 6

7 CAN When a peer leaves, a takeover algorithm updated the allocation of peer in zone. The number of neighbor a peer maintains depends only on the dimensionality of the coordination space (i.e. 2 x d) and it is independent of the total number of peers in the system 13 CAN Improvement of CAN Reality Multiple, independent coordinate spaces A CAN with r realities, a single peer is assigned r coordinate zones with r independent neighbor sets The content of the hash table is replicated on each reality, improving data availability K different hash functions to map a given key onto k points in teh coordinate space This is a replication of a single {key,value} pair at k distinct peers in the system Queries for a particular hash table entry could be forwarded to all k peers in parallel, reducing the average query latency Reliability and fault resiliency properties are enhanced too 14 7

8 CAN Examples of application of CAN OceanStore architecture for global-scale persistent storage (2002) Farsite Serverless distribute file system deployed on a existing set of desktop pcs (2000) Publius Robust, censorship-resistant web publishing (2000) Construction of a wide-area name resolution services that decople the naming scheme for the name resolution process. This enables an arbitrary and locationindependent naming scheme 15 Chord Consistenf hashing to assign keys to its peers. Peers enter and leave the network with minimal interruption Tends to balance the load on the system, since each peer receives roughly the same number of keys, and there is a little movement of keys when peers join and leave the system Routing performance ofθ(d x N 1/d ) Routing stateθ( log N) 16 8

9 Chord The consistent hash function assign peers and data keys an m-identifier using SHA-1 Peer identifier = H (peer`s IP address) Key identifier = H (data key) The lenght of m identifier must be large enough Identifiers are ordered on a circle modulo 2m (from 0 to 2m-1). Chord Ring Key k is assigned to the first peer whose identifier is equal to or follows k in the identifies space. Sucessor (k) Sucessor (k) is the first peer clockwise from k 17 Chord When a peer n joins the network, certain keys previously assigned to n successors now need to be reasigned to n. When peer n leaves the network, all of its assigned keys are reassigned to n s sucessor. Peers join and leave the network (logn) 2 performance No other changes on key assignment needed 18 9

10 Chord Fuente: [Survey05] 19 Chord Chord ring m= 6 10 peers Store 5 keys The sucessor of the identifier 10 is peer 14, so key 10 will be located at NodeID 14 If a peer were to join with identifier 26, it would store the key with identifier 24 from the peer with identifier 32 Each peer maintains a routing table with up to m entries: finger table (chord id, IP) 20 10

11 Chord Peers store information only of a few number of other peers, and know more about peer closely following it on the circle. Also, a peer s finger table does not contain enough information to directly determine the sucessor of an arbitrary key k Whe a peer join, the sucessor pointers need to be updated by a stabilization protocol running periodically. Each peer is aware of its sucessor When a peer fail, it is possible that a peer does not know its new sucessor, and it has no chance to learn about it. To avoid this, each peer maintains a successor lists of r peers whic contains the peer s first r sucessors If a successor peer doesn t respond, the peer contacts the next peer in the sucessor list. Increasing r makes the system more robust 21 Chord Examples of applications of Chord: Cooperative mirroring or coorperative files systems (CFS). Multiple providers of content cooperate to store and serve each other s data Each participant need to provide capacity only for the average load, not for the peak load Chord-base DNS Lookup service Host name -> keys IP address -> values Rely on root servers not needed Advantages.. > work proposal 22 11

12 Structured CAN (2000) Chord (2003) Tapestry (2004) Pastry (2001) Kademlia (2002) Viceroy (2002) 23 Discussion about Structured p2p DHT-based systems have a few problems in terms of data object lookup latency: For each overlay hop, peers route a message to the next intermediate peer that can be located very far away with regard to physical topology or the underlying IP network It is assumed that all peers equally participate in hosting published data objects or their location information. Bottleneck of low-capacity peers DHT-based systems do not capture the semantic object relationship between name and its content or metadata Not widely deployed 24 12

13 Discussion about Structured p2p Security A malicious peer return wrong data objects to the lookup queries Criptographic techniques for data authenticity.. NodeID assignment Peer selection in routing Cross checking using random queries Avoiding single points of responsability Taxonomy of attacks -> work proposal Analysis of techniques for secure peer joining, routing table maintenance, robust message forwarding in the case of a set of malicious peers (Eclipse attack) 25 Unstructured Overlay networks organize peers in a random graph in a flat or hierarchical manner (ie Super-Peers layer) Use flooding or random walks or expanding-ring TTL search Each peer will support complex queries Inefficient becaus queries for content that are not widely replicated must be sent to a large fraction of peers There is no coupling between topology and data item s location 26 13

14 Unstructured Freenet Gnutella FastTrack/KazaA BitTorrent Work proposal.- Description of more common unstructured p2p overlay networks 27 Discussion of Unstructured p2p More efficient in a mass-market file sharing Because DHT-based solutions are not widely deployed, researh work is trying to improve unstructured solutions to include flow control, dynamic geometric topology adaptation, one-hop replication, peer heterogeneity 28 14

15 Future in p2p research How the p2p virtual topology maps in to the physical network infrastructure Quantitative evaluation on p2p overlay applications and Internet topology matching Scalability of p2p by the efficient use of the undelying physical network resources Self-regulatory auditing and accounting behaviour for resource sharing Economic and game theories for p2p collaboration is crutial to create a economy of equilibrium Trust and reputation Cross-application of Internet p2p overlay networking models in mobile, wireless, ad-hoc networks 29 Applications Voip in p2p Skype DHT-based solutions p2psip 30 15

16 p2p and SIP INVITE REGISTER => user_a at peer_1 Contact: uc3m.es user_b at peer_ Client-serve Model=> maintenance, configuration, controlled infrastructure INVITE user_b P2P overlay REGISTER user_b P2P Model => No central server (search latency) 31 p2p and SIP SIP-using-P2P, replace SIP location service by a p2p protocol P2P-over-SIP, implementation of p2p using SIP signaling FIND P2P network INVITE sip:user_b@ INSERT user_b INVITE user_b P2P-SIP overlay REGISTER user_b

17 SIP-using-P2P Reusing optimized and well-defined external P2P network Defining P2P location service interface to be used in SIP Extends to other signaling protocols 33 P2P-over-SIP P2P algorithm over SIP without change in semantics No dependence on external P2P network Study how to built-in NAT/media relays 34 17

18 P2P-SIP scenarios Non server-based architecture Non propietary architecture (skype) Robust and efficient lookup using DHT DHT can use SIP communication Hybrid scenario: lookup SIP and DHT Applications: use p2p-sip as an outbound proxy P2P vs server-based SIP P2P for smaller scenarios: community network Server-based for bigger scenarios: carriers Federated systems: multiple p2p systems, identified by DNS domain name with gateway nodes 35 P2P-SIP scenarios P2P vs server-based SIP P2P for smaller scenarios: community network Server-based for bigger scenarios: carriers Federated systems: multiple p2p systems, identified by DNS domain name with gateway nodes P Q R S server based company.com E A G D content.eu H uc3m.es C F B P2P-SIP nodes X uc.pt Z Y P2P-SIP nodes 36 18

19 Standardization IRTF P2P Research Group prg P2P-SIP A P2P Approach to SIP Registration (draft) 37 Bibliografía [Survey05] A survey and comparision of peer-to-peer overlay network schemes Eng Keong Lua, J. Crowcroft and M. Pias. IEEE Communications Survey, 2nd Quarter Vol 7. No. 2 Peer-to-Peer Systems and Applications R. Steinmetz and K. Wehrle. State of the art survey. LNCS Springer

20 Trabajos propuestos p2psip. Análisis de la arquitectura Aspectos de seguridad Ejemplos de KBR APIs: OpenDHT (open public accesible DHT service) Soluciones de p2p para movilidad de redes Infraestructura para descubrimiento de HA Servicios básicos de Internet en p2p: Chord-based DNS What p2p-based DNS? Análisis de seguridad en DHT-based p2p Taxonomy of attacks Description of more common unstructured p2p overlay networks 39 20