Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

31
Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper

Transcript of Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Page 1: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Impact of Incentives in BitTorrent

By Jenny Liu and Seth Cooper

Page 2: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Peer to Peer Networks

• Multiple independent peers connect to share resources with each other

Page 3: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Incentives

• But what reason does one peer have to share its resources with another?

Page 4: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Incentives

• Does a peer need something in return?

Page 5: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Incentives

• Or is simple altruism enough?

Page 6: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

BitTorrent

• Torrent file directs peers to a tracker, which helps to coordinate peers

Torrent

Tracker

Page 7: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

BitTorrent

• Initial peer, known as a “seed,” begins with the complete file.

Torrent

Tracker Seed

Page 8: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

BitTorrent

• Peers trade pieces of the file amongst themselves, uploading pieces to the peers they are downloading the fastest from. When a peer finishes, it becomes a new seed.

Torrent

Tracker Seed

Page 9: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Motivation

• Other P2P networks (Napster, Kazaa) have become successful relying on altruism alone.

Page 10: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Motivation

• It is often claimed that freeloading on P2P networks is a problem, and many attempts have been made to find an effective incentive, but is one necessary?

Page 11: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Motivation

• How effective is BitTorrent? Are incentives necessary? How do greedy peers affect the network? Are incentives an unnecessary overheard?

Page 12: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

System

• Modify BitTorrent source to remove incentives, and compare efficiency with unmodified BitTorrent

Page 13: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

System

• Assign each peer a strategy: greedy or altruistic.

Page 14: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Metrics

• Client side: average download time

Page 15: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Metrics

• Server side: percent of unicast

Page 16: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Trials

• Measure data from several trials of one seed distributing a file to a set of peers. Start with a default network setting and vary parameters to see their effect on the two types of networks.

Seed

Page 17: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Trials

• We varied: delay between peer connections.

Page 18: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Trials

• We varied: minimum number of uploads for each peer.

Page 19: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Results

0

5

10

15

20

25

30

35

40

45

50

0 2 4 6 8 10

Tim

e to

com

plet

e do

wnl

oad

Delay coefficient

Effect of delay coefficient on time to complete download

vanillano incentive

0

20

40

60

80

100

0 2 4 6 8 10

Per

cent

of u

nica

st

Delay coefficient

Effect of delay coefficient on percent of unicast

vanillano incentive

ideal

0

5

10

15

20

25

30

35

40

45

50

1 2 3 4 5 6 7Ti

me

to c

ompl

ete

dow

nloa

dMinimum number of uploads

Effect of minimum number of uploads on time to complete download

vanillano incentive

0

20

40

60

80

100

1 2 3 4 5 6 7

Per

cent

of u

nica

st

Minimum number of uploads

Effect of minimum number of uploads on percent of unicast

vanillano incentive

ideal

• Not much difference.

Page 20: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Trials

• We varied: total number of peers.

Page 21: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Results

0

5

10

15

20

25

30

35

40

45

50

2 4 6 8 10 12 14 16

Tim

e to

com

plet

e do

wnl

oad

Number of peers

Effect of number of peers on time to complete download

vanillano incentive

• Number of peers has an effect on download time when the incentive is removed.

Page 22: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Trials

• We varied: percentage of greedy peers.

Page 23: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Results

0

20

40

60

80

100

0 20 40 60 80 100

Per

cent

of u

nica

st

Percentage of greedy peers

Effect of percentage of greedy peers on percent of unicast

vanillano incentive

ideal

• Percentage of greedy peers has an effect on both metrics when the incentive is removed.

0

50

100

150

200

0 20 40 60 80 100

Tim

e to

com

plet

e do

wnl

oad

Percentage of greedy peers

Effect of percentage of greedy peers on time to complete download

vanillano incentive

Page 24: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Trials

• We varied: size of network and tried different percentages of greedy hosts

Page 25: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Results

• As the network gets bigger, a larger percentage of greedy hosts can be supported

0

50

100

150

200

0 20 40 60 80 100

Tim

e to

com

plet

e do

wnl

oad

Percentage of greedy peers

Effect of percentage of greedy peers on time to complete download

vanillano incentive

0

50

100

150

200

0 20 40 60 80 100Ti

m to

com

plet

e do

wnl

oad

Percent of greedy peers

Effect of percent of greedy peers on time to complete download

vanillano incentive

10 Peers 20 Peers

Page 26: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Conclusions

• Q: How efficient is BitTorrent?

• A: Very.

Page 27: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Conclusions

• Q: Are incentives necessary?

• A: It depends.

Page 28: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Conclusions

• Q: How do greedy peers affect the network?

• A1: Not much. • A2: Not much up to a point. • A3: Quite a lot after that.

Page 29: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Conclusions

• Q: Are incentives an unnecessary overhead?

A: No!

Page 30: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

Future Work

• Conduct experiments on more heterogeneous networks.

• Try UDP.

• Address other negative aspects of BitTorrent, such as content lifetime.

Page 31: Impact of Incentives in BitTorrent By Jenny Liu and Seth Cooper.

The End