1. CSE 461 University of Washington1 Topic Peer-to-peer content delivery – Runs without dedicated infrastructure – BitTorrent as an example Peer

2. CSE 461 University of Washington2 Context Delivery with client/server CDNs: – Efficient, scales up for popular content – Reliable, managed for good service … but some disadvantages too: – Need for dedicated infrastructure – Centralized control/oversight

3. CSE 461 University of Washington3 P2P (Peer-to-Peer) Goal is delivery without dedicated infrastructure or centralized control – Still efficient at scale, and reliable Key idea is to have participants (or peers) help themselves – Initially Napster ‘99 for music (gone) – Now BitTorrent ‘01 onwards (popular!)

4. CSE 461 University of Washington4 P2P Challenges No servers on which to rely – Communication must be peer-to-peer and self-organizing, not client-server – Leads to several issues at scale … Peer

5. CSE 461 University of Washington5 P2P Challenges (2) 1.Limited capabilities – How can one peer deliver content to all other peers? 2.Participation incentives – Why will peers help each other? 3.Decentralization – How will peers find content?

6. CSE 461 University of Washington6 Overcoming Limited Capabilities Peer can send content to all other peers using a distribution tree – Typically done with replicas over time – Self-scaling capacity Source

7. CSE 461 University of Washington7 Overcoming Limited Capabilities (2) Peer can send content to all other peers using a distribution tree – Typically done with replicas over time – Self-scaling capacity Source

8. CSE 461 University of Washington 8 Providing Participation Incentives Peer play two roles: – Download ( ) to help themselves, and upload ( ) to help others Source

9. CSE 461 University of Washington9 Providing Participation Incentives (2) Couple the two roles: – I’ll upload for you if you upload for me – Encourages cooperation Source

10. CSE 461 University of Washington10 Enabling Decentralization Peer must learn where to get content – Use DHTs (Distributed Hash Tables) DHTs are fully-decentralized, efficient algorithms for a distributed index – Index is spread across all peers – Index lists peers to contact for content – Any peer can lookup the index – Started as academic work in 2001

11. CSE 461 University of Washington11 BitTorrent Main P2P system in use today – Developed by Cohen in ‘01 – Very rapid growth, large transfers – Much of the Internet traffic today! – Used for legal and illegal content Delivers data using “torrents”: – Transfers files in pieces for parallelism – Notable for treatment of incentives – Tracker or decentralized index (DHT) By Jacob Appelbaum, CC-BY-SA-2.0, from Wikimedia Commons Bram Cohen (1975—)

12. CSE 461 University of Washington12 BitTorrent Protocol Steps to download a torrent: 1.Start with torrent description 2.Contact tracker to join and get list of peers (with at least seed peer) 2.Or, use DHT index for peers 3.Trade pieces with different peers 4.Favor peers that upload to you rapidly; “choke” peers that don’t by slowing your upload to them

13. BitTorrent Protocol (2) All peers (except seed) retrieve torrent at the same time CSE 461 University of Washington13

14. BitTorrent Protocol (3) Dividing file into pieces gives parallelism for speed CSE 461 University of Washington14

15. BitTorrent Protocol (4) Choking unhelpful peers encourages participation CSE 461 University of Washington15 STOP XXX

16. BitTorrent Protocol (5) DHT index (spread over peers) is fully decentralized CSE 461 University of Washington16 DHT

17. CSE 461 University of Washington17 P2P Outlook Alternative to CDN-style client- server content distribution – With potential advantages P2P and DHT technologies finding more widespread use over time – E.g., part of skype, Amazon – Expect hybrid systems in the future

18. Chord: A Scalable Peer-to-peer Lookup Service for Internet Applications Robert Morris Ion Stoica, David Karger, M. Frans Kaashoek, Hari Balakrishnan MIT and Berkeley

19. A peer-to-peer storage problem 1000 scattered music enthusiasts Willing to store and serve replicas How do you find the data?

20. The lookup problem Internet N1N1 N2N2 N3N3 N6N6 N5N5 N4N4 Publisher Key= “ title ” Value=MP3 data… Client Lookup( “ title ” ) ?

21. Centralized lookup (Napster) Publisher@ Client Lookup( “ title ” ) N6N6 N9N9 N7N7 DB N8N8 N3N3 N2N2 N1N1 SetLoc( “ title ”, N4) Simple, but O( N ) state and a single point of failure Key= “ title ” Value=MP3 data… N4N4

22. Flooded queries (Gnutella) N4N4 Publisher@ Client N6N6 N9N9 N7N7 N8N8 N3N3 N2N2 N1N1 Robust, but worst case O( N ) messages per lookup Key= “ title ” Value=MP3 data… Lookup( “ title ” )

23. Routed queries (Freenet, Chord, etc.) N4N4 Publisher Client N6N6 N9N9 N7N7 N8N8 N3N3 N2N2 N1N1 Lookup( “ title ” ) Key= “ title ” Value=MP3 data…

24. Routing challenges Define a useful key nearness metric Keep the hop count small Keep the tables small Stay robust despite rapid change Freenet: emphasizes anonymity Chord: emphasizes efficiency and simplicity

25. Chord properties Efficient: O( log(N) ) messages per lookup – N is the total number of servers Scalable: O( log(N) ) state per node Robust: survives massive failures Proofs are in paper / tech report – Assuming no malicious participants

26. Chord overview Provides peer-to-peer hash lookup: – Lookup(key)  IP address – Chord does not store the data How does Chord route lookups? How does Chord maintain routing tables?

27. Chord IDs Key identifier = SHA-1(key) Node identifier = SHA-1(IP address) Both are uniformly distributed Both exist in the same ID space How to map key IDs to node IDs?

28. Consistent hashing [Karger 97] N32 N90 N105 K80 K20 K5 Circular 7-bit ID space Key 5 Node 105 A key is stored at its successor: node with next higher ID

29. Consistent hashing [Karger 97] Theorem: For any set of N nodes and K keys, with “high probability” 1)Each node is responsible for at most (1+ eps) K/N keys 2)When the (N+1)th node joins or leaves the network, responsibility for O(K/N) keys changes hands

30. Basic lookup N32 N90 N105 N60 N10 N120 K80 “ Where is key 80? ” “ N90 has K80 ”

31. Simple lookup algorithm Lookup(my-id, key-id) n = my successor if my-id < n < key-id call Lookup(id) on node n // next hop else return my successor // done Correctness depends only on successors

32. “ Finger table ” allows log(N)-time lookups N80 ½ ¼ 1/8 1/16 1/32 1/64 1/128

33. Finger i points to successor of n+2 i N80 ½ ¼ 1/8 1/16 1/32 1/64 1/128 112 N120

34. Lookup with fingers Lookup(my-id, key-id) look in local finger table for highest node n s.t. my-id < n < key-id if n exists call Lookup(id) on node n // next hop else return my successor // done

35. Lookups take O( log(N) ) hops N32 N10 N5 N20 N110 N99 N80 N60 Lookup(K19) K19

36. Failures might cause incorrect lookup N120 N113 N102 N80 N85 N80 doesn ’ t know correct successor, so incorrect lookup N10 Lookup(90)

37. Solution: successor lists Each node knows r immediate successors After failure, will know first live successor Correct successors guarantee correct lookups Guarantee is with some probability

38. Choosing the successor list length Assume 1/2 of nodes fail P(successor list all dead) = (1/2) r – I.e. P(this node breaks the Chord ring) – Depends on independent failure P(no broken nodes) = (1 – (1/2) r ) N – r = 2log(N) makes prob. = 1 – 1/N