Free-riding in BitTorrent Networks with the Large View

FREE-RIDING IN BITTORRENT
NETWORKS WITH THE LARGE
VIEW EXPLOIT
Michael Sirivianos, et al, University of California, Irvine, IPTPS 2007
2009. 12. 5
Presented by Jinyoung Han
* Presentation material is borrowed from user’s
CONTENTS
Introduction
 Free-riding in BitTorrent
 Large view exploit
 Experimental evaluation
 Conclusion

2
WHO IS A FREE-RIDER?

The name "free-rider" comes from a common text
book example:


Someone using public transportation without paying
the fare. If too many people do this, the system will n
ot have enough money to operate [1]
Free-rider in P2P system

Selfish users tend not to share their bandwidth


They try to just download only and not to upload anything
85% of the peers do not share any file in Gunutella [2]
[1] wikipedia.org
[2] D. Hughes, et al, “Free riding on gnutella revisited: The bell tolls?”,
IEEE Distributed Systems Online, June 2005
3
BITTORRENT

Very successful P2P content distribution system

Built-in incentives for clients to upload
Positively correlates a client’s download and upload
rates
 Tit-for-tat


However, BitTorrent users may still choose to
free-ride
If they can download as fast as compliant clients…
 Limited uplink bandwidth
 Less risk of legal action

4
HOW BITTORRENT WORKS (1)
View
Partial View
Leecher A
Seeder
Tracker
Leecher B
Seeder
Leecher A
Leecher B
Seeder
Leecher C
Leecher C
Leecher B

Clients contact the tracker



Tracker has complete view of the swarm
Tracker sends partial view to clients
Clients connect to peers in their partial view
5
HOW BITTORRENT WORKS (2)
Tracker
Leecher A
Seeder
Leecher C
Leecher B

Seeder sample their peers’ download rates



Seeder unchoke 5 interested fastest downloaders
Clients announce available chunks to their peers
Leechers request chunks from their peers (locally rarest-first)
6
HOW BITTORRENT WORKS (3)
Leecher A
Seeder
Tracker
Leecher C
Leecher B

Rate-based tit-for-tat


Leecher optimistically unchoke 1 peer
Leecher sample their peers’ upload rates and unchoke 4
fastest interested uploaders
7
FREE-RIDING IN BITTORRENT

Observation
Seeder does not consider whether leechers upload to
others
 Leecher’s optimistic unchoking is generous


Even if a client does not upload, it may be able to
download faster than tit-for-tat compliant clients

Especially in a large swarm
Increasing possibility to be optimistically unchoked
 Connecting to more seeders

8
LARGE VIEW EXPLOIT

Modify BitTorrent to exploit altruism
It never uploads to its peers
 Connect from and disconnect to tracker every 15 sec



Repeatedly request and obtain partial views
Connect to all clients in the combined large view
It is more likely to connect to seeders
 It is more likely to be optimistically unchoked by leechers

9
EXPERIMENTAL EVALUATION

Goals
Determine effectiveness of the large view exploit
 Determine behavior of swarms as % free-riders
increases


Performance metric


File download completion time
Two scenarios
In public swarms
 In planetlab-residing swarms

10
FREE-RIDING IN PUBLIC SWARMS (1)

A CTorrent free-rider and a
compliant (unmodified)
leecher
Compliant leecher connects
to at most 50 peers
 Run on two identical
machines and in the same
LAN
 Join to the swarm
simultaneously

11
FREE-RIDING IN PUBLIC SWARMS (2)
Swarm

#seeders
#leechers
13
9
64
14
91
392
In 12/15 swarms, free-rider does better than
compliant client
12
FREE-RIDING IN PLANETLAB-RESIDING
TORRENTS (1)

CTorrent free-riders and compliant (unmodified)
leechers
Compliant leechers connect to at most 50 peers
 Free-rider obtain a swarm view of ~ 250 clients


Experiment settings
Item
Value
File size
12MB
# of leechers
(both free-riders and compliants)
~300
# of seeder
1
Rate-limit of leechers
30KB/sec
Rate-limit of seeder
120KB/sec
13
FREE-RIDING IN PLANETLAB-RESIDING
TORRENTS (2)

Effectiveness of the large view exploit
[ Without large view exploit ]
[ With large view exploit ]
14
FREE-RIDING IN PLANETLAB-RESIDING
TORRENTS (3)

Behavior of swarms as % free-riders increases
15
CONCLUSION

The large view exploit is effective to free-rider


New aspect of the free-riding problem in BitTorrent
Selfish users may opt to free-ride
A free-rider can achieve better or similar download
rates than compliant clients
 However, bad download rates when many free-riders
in swarm

16