Chapter 12: P2P Applications

Question 1

Spotify.com, 2004

Answer 1

• Commercially deployed system, KTU start-up (Sweden)
• Peer-assisted on-demand music streaming
• Legal and licensed content only
• Large catalogue of music (over 15 million tracks)
• Available in U.S. & 7 European countries, over 10 million users, 1.6 million subscribers
• Fast (median playback latency of 265 ms)
• Proprietary client software for desktop & phone (not p2p)
• Business model: Ad-funded and free & monthly subscription, no ads, premium content, higher quality streaming

Question 2

Overview of Spotify Protocol

Answer 2

• Proprietary protocol
• Designed for on-demand music streaming (share to peers)
• Only Spotify can add tracks
• 96–320 kbps audio streams (most are Ogg Vorbis q5, 160 kbps)
• Relatively simple and straightforward design

Question 3

Why a peer-to-peer protocol? Spotify

Answer 3

• Improve scalability of service
• Decrease load on servers and network resources
• Explicit design goal:
  
  • Use of peer-to-peer should not decrease overall performance (i.e., playback latency & stutter)

Question 4

Spotify P2P overlay structure

Answer 4

• Nodes have fixed maximum degree (60)
• Neighbour eviction by heuristic evaluation of utility
• ( multi objective [Pound , Upload ( opacity] ) ks Utility function
• Limited overlay routing (search requests)
• Looks for and connects to new peers when streaming new track
• Overlay becomes (weakly) clustered by interest
• Client only downloads data user needs

Question 5

Finding peers Spotify

Answer 5

• Sever-side tracker (cf. BitTorrent)
  
  • Only remembers 20 peers per track
  • Returns 10 (online) peers to client on query
• Clients broadcast query in small (2 hops) neighbourhood in overlay (cf. Gnutella)
• Client uses both mechanisms for every track

Question 6

Spotify Protocol

Answer 6

• (Almost) everything encrypted
• (Almost) everything over TCP
• Persistent connection to server while logged in
• Multiplex messages over a single TCP connection

Question 7

Spotify Caches

Answer 7

• Client (player) caches tracks it has played
• Default policy is to use 10% of free space (capped at 10 GB)
• Caches are often larger (56% are over 5 GB)
• Least Recently Used policy for cache eviction hk
• Over 50% of data comes from local cache
• Cached files are served in peer-to-peer overlay (if track completely downloaded)

Question 8

Spotify - Streaming a Track

Answer 8

• Tracks are decomposed into 16 kB chunks
• Request first piece of track from Spotify servers
• Meanwhile, search for peers that cache track
• Download data in-order (chunk by chunk via TCP)
• Ask for most urgent pieces first
• Towards end of a track, start prefetching next track
• If a peer is slow, re-request data from new peers
• If local buffer is sufficiently filled, only download from peer-to-peer overlay play
• If buffer is getting low, stop uploading

Question 9

Spotify Security through obscurity

Answer 9

• Music data lies encrypted in caches
• Client must be able to access music data
• Reverse engineers should not be able to access music data
• Details are secret and client code is obfuscated
• Security through obscurity is a bad idea

Question 10

Spotify Key points

Answer 10

• Simplicity of architecture, protocol, design
• Peer-assisted, i.e., rely on centralized server
• Use of peer-to-peer techniques for scalability and avoid heavy, over-provisioned infrastructure
• Use of centralized tracker

Question 11

BitTorrent High Level

Answer 11

• Pull-based, swarming approach
  
  • Each file is split into smaller pieces (& sub-pieces)
  • Peers request desired pieces from neighboring peers
  • Pieces are not downloaded in sequential order
• Encourages contribution by all peers
• Based on a tit-for-tat model

Question 12

BitTorrent use cases

Answer 12

• File-sharing
• What uses does BitTorrent support?
  
  • Downloading (licensed only ) movies, music, etc.
• Update distribution among servers at Facebook et al.
• Distribution of updates and releases (e.g., World of War.)
• •…

Question 13

BitTorrent Swarm

Answer 13

• Swarm
  
  • Set of peers downloading the same file
  • Organized as a randomly connected mesh of peers
• Each peer knows list of pieces downloaded by neighboring peers
• Peer requests pieces it does not own from neighbors

Question 14

BitTorrent terminology

Answer 14

• Seed: Peer with the entire file
  
  • Original Seed: First seed for a file
• Leech: Peer downloading the file
  
  • Leech becomes a seed, once file downloaded, if the peer stays online & continues by protocol
• Sub-piece: Further subdivision of a piece
  
  • “Unit for requests” is a sub-piece (16 kB)
  • Peer uploads piece only after assembling it completely

Question 15

How a node enters a swarm for file “popeye.mp4” Bitorrent

Answer 15

• File popeye.mp4.torrent hosted at a (well- known) web server
• The .torrent has address of tracker for file
• The tracker, which runs on a web server as well, keeps track of all peers downloading file

Question 16

Contents of .torrent file

Answer 16

• URL of tracker
• Piece length – Usually 256 KB
• SHA-1 hashes of each piece in file - Why?
• Hash helps to check if file downloaded is the right one

Question 17

Bitorrent 3 Phases

Answer 17

• Beginning: First piece
  
  • First, pick a random piece (random first policy)
    • When peer starts to download, request random piece
    • When first complete piece assembled, switch to rarest- first
    • Get first piece to quickly participate in swarm with upload
    • Rare pieces are only available at few peers (slows
• Middle: Piece 1 to n-1
  
  • Strict priority policy
    • Always complete download of piece (sub-pieces) before starting a new piece
    • Getting a complete piece as quickly as possible
• End-game: Piece n
  
  • When requests sent for all sub-pieces of last

piece, re-send requests to all peers

* **Upon** **download** of **entire** **piece**, **cancel** **request** for downloading **sub**-**pieces** **from** **other** **peers**
* **Speeds** **up** **completion** of download, **otherwise last piece could delay completion of download**

Question 18

Tit-for-tat as incentive to upload ​Bitorrent

Answer 18

• Want to encourage all peers to contribute
• Peer A said to choke peer B if A decides not to upload to B
• Each peer (say A) unchokes at most 4 interested peers at any time
  
  • The three with the largest download rates from A
    
    • Where the tit-for-tat comes in
  • Another randomly chosen (optimistic unchoke)
    
    • Periodically discover better choices

Question 19

Why BitTorrent took off

Answer 19

• Working implementation (Bram Cohen) with simple well- defined interfaces for publishing content
• Open specification
• Many competitors got sued & shut down (Napster, KaZaA)
• Simple approach
  
  • Doesn’t do “search” per se
  • Users use well-known, trusted sources to locate content

Question 20

Pros & cons of BitTorrent

Answer 20

Pros:

• Proficient in utilizing partially downloaded files
• Discourages“freeloading”
  
  • By rewarding fastest uploaders
• Encourages diversity through “rarest-first”
  
  • Extends lifetime of swarm
• Works well for “hot content”

​Cons:

• Assumes all interested peers active at same time
• Performance deteriorates if swarm “cools off”
• Too much overhead to disseminate small files

Question 21

Spotify vs. BitTorrent

Answer 21

Spotify:

• Live listening of streamed track
• One peer-to-peer overlay for all tracks
• Does not inform peers about downloaded blocks
• Downloads blocks in order
• Does not enforce fairness
• Informs peers about urgency of request
• Supports search

Bitorrent:

• Batch download of large files
• Essentially, a swarm (overlay) per torrent
• Exchange of downloaded blocks with peers
• Random download order
• Tit-for-tat (game theoretic roots)
• No notion of urgency
• No search feature

Question 22

Skype network

Answer 22

• Super Nodes: Any node with a public IP address having sufficient CPU, memory and network bandwidth is candidate to become a super node C know who is online)
• Ordinary Host: Host needs to connect to super node and must register itself with the Skype login server
• Login server and PSTN gateway (not shown) are only centralized components

Question 23

Impact of Skype

Answer 23

• Skype has shown, at least has suggested, the following
  
  • Signaling, the most unique property of traditional phone systems, can now be accomplished effortlessly with self-organizing P2P networks
  • P2P overlay networks can scale up to handle large- scale connection-oriented real-time services such as voice

Question 24

Common principles

Answer 24

• Peer-assisted, hybrid architecture
• Use of tracker, as directory (centralized)
• Peer-to-peer capability to limit infrastructure investments & achieve scalability
• Use of “heavier” TCP, due to inherent guarantees
• Simplicity of design, as simple as possible