1
Take One Get One Free Leveraging P2P Networks
for Content Promotion

• INFOCOM06

2
Outline

• Introduction
• Promoter
• Architectural Components
• Experiment
• Discussion

3
Introduction to Current Status

• Traditionally, digital content is primarily
  produced by commercial web sites to be downloaded
  by users
• Now comes an explosion of user-generated content.
• They have a strong motivation to promote their
  content
• However, both WWW and P2P play a passive role
  waiting for the seekers to discover its content
  in a pull-based means.

4
Introduction to This Paper

• This paper advocates an active push-based
  mechanism to promote content as a complement to
  the existing pull-based content sharing
• Promoter is proposed to publicize the existence
  of content among other users, and attempt to
  disseminate the content to as many peers as
  possible.

5
Promoter

• It builds upon existing P2P file sharing protocol
  that provide search and reply
• It enables two new primitives psearch and preply
  
• Let A be a peer issuing the search K consisting
  of n keywords ( Kk1, k2,kn ) and B be a peer
  receiving the search

6
Publicizing

• Peers separate content into promote-worthy
  content (P) and general content (G)
• If A is interested in learning about promoted
  content, it issues a psearch identical to search,
  except for a flag indicating its interest
• If B doesnt have matching content, or its P is
  empty, or content matching K is part of P in B, B
  behaves as if it received a basic search message.
  Hence, no publicizing is attempted
• If content matching K is in G, B generates preply
  which piggybacks infos about content from P, in
  addition to search hit infos. Each hit is
  accompanied with one promoted item info
• A is informed of promoted content of all the
  peers that hit K
• Users can disable this function and use search
  only

7
Disseminating

• B disseminates its P as follows
• When B has a hit in P, it piggybacks a download
  incentive offer in preply that rewards A if it
  chooses Bs content over other peers
• When B has a hit in G, it piggybacks a take one
  get one free offer along with a download
  incentive to A.
• If A accepts, it downloads these two shares of
  content and collects the associated reward.
• B incentivizes peers to
• 1) download P
• 2) offer it to others after download
• 3) promote it after download

8

• The download incentive allows promoted item to be
  put in G of other peers, or even P by providing
  propagate incentive.
• The associated rewards for propagate incentive
  are collected when A provides proof to B that it
  propagated the content further to new peer

9
Design of Incentives

• Currently, P2P systems incentivize to avoid
  free-riding problems
• Difference of Promoters incentives from P2P
• to motivate peers to download, not force them in
  order to survive
• It is local (between pairs of peers), while some
  P2Ps incentives like reputation are typically
  system-wide and long term.

10
Download Propagate Incentives

• DI motivates users to download promoted items and
  should have immediate gratification.
• The promoting peer offers high QoS by reserving
  an advertised amount of upload bandwidth from its
  end for the receiving peer throughout the
  download of the queried content and the promoted
  content.
• PI, by its nature, has delayed gratification.
• The promoting peer offers a bandwidth reservation
  for one future download to the receiving peer

11
Architectural Components

• The promoting peer requires the Publicizer, Offer
  Generator, and Connection Manager modules.
• The receiving peer requires the Offer Acceptor
  and Collector, and QoS Monitor modules.

12
Publicizer Module (PM)

• PM handles all the protocol messages and
  communicates with other modules to enable content
  promotion.
• PM exports an interface for the user to separate
  shared content into P and G categories.
• Each item I in P has the following metadata A
  user-assigned rank Boolean value to specify
  whether DI and/or PI will be provided and a token
  list
• It operates as follows

13

• Search in PG for keywords received in psearch
• If a hit H belongs to P and its DI flag is set,
  PM piggybacks a download incentive offer from OGM
  in preply
• If H belongs to G, PM first picks an item I from
  P to promote such that the number of times an
  item is chosen is proportional to its rank and
  its size doesnt exceed H
• Repeat this process for each hit item for
  keywords in psearch
• These item pairs are included in preply along
  with a description of each I
• Requests OGM for a bandwidth reservation for
  preply
• Send and await accept and conveys its receipt to
  CMM and OGM
• PM also sends appropriate propagate incentives to
  the peers that download a promoted item

14
Offer Generator Module (OGM)

• OGM provides download incentive, offering
  bandwidth reservation to PM to be sent to the
  querying peer
• At bootstrap, the user sets aside UB, a
  configurable fraction of the total upload
  bandwidth, for making offers and the OGM
  dynamically manages UB.
• An offer essentially reserves a fraction of UB
  for a peer.
• A trivial method reserve entire UB for a single
  offer and reuse ? underutilization due to
  querying peer download bandwidth constraints or
  network properties
• Potential solutions
• Parallel downloads for different byte ranges of
  the same content for certain conditions
• Predict/Monitor achievable throughputs of all the
  current connections and use surplus for future
  offer

15
Connection Manger Module (CMM)

• CMM enforces the QoS offers made to peers, in
  addition to aiding the OGM in thwarting
  underutilization

16
Offer Acceptor and Collector Module (OACM)

• After receiving all the replies, choose
  automatically or by user greedy approach /
  download different byte ranges of the content
  from multiple peers
• Propagate incentive
• OACM stores for each item I in P the list of
  accepted PI offer tokens, each of which contains
  the offering peers IP and an offer id
• Note that A can even choose to provide its own
  propagate offer tokens for promoting I to C

17
QoS Monitor Module (QMM)

• QMM monitors the QoS offered connections to check
  if the promoter is respecting the offers made or
  not.
• Note that when an offer is flagged as violated,
  the QMM needs to ensure that the limitation is
  not on its end

18
Experiment Settings

• 1540 peers, 8019 files, 26959 keywords, 20000
  queries
• Random upload bandwidth in 300kbps, 600kbps
• Query is flooded via its 4 randomly chosen
  neighbors with a TTL of 4
• Every peer has promote-worthy content and
  piggybacks one item for every file returned in a
  hit
• The peer reserves 70 of its upload bandwidth as
  a download incentive
• The querying peer greedily choose the content
  item with the highest reserved bandwidth
• The size of each file in 1MB, 10MB

19
Experiment Result

• Publicity Measure number of times a node is able
  to publicize its content.
• It depends on total number of queries in system,
  connectivity of the querying nodes, and
  popularity of files at the nodes receiving the
  query
• Competitive publicizing peers monitor forwarded
  messages and crawl k popular files. K20, 100,
  500

20
Experiment Result

• Dissemination Measure number of times a node is
  able to disseminate promote-worthy content.
• It depends on its PM and its upload bandwidth
  compared to other peers responding to a query
• Note that these results are conservative since
  they ignore the downloads that may occur due to
  interest generated from publicizing as well as
  from propagate incentives.

21
Discussion

• Free riding and Promoting
• Although free-riders may not allow uploads
  of general content, it is likely that a fraction
  of free-riders will have a passion for promoting
  their own content. In this scenario, Promoter may
  reduce free-riding by forcing such users to
  upload general content in order to promote their
  own.
• Future Work
• Promotion filter (similar to spam filter)
• Use keywords in the search message to pick an
  item to be promoted

22
Thank You !