Connectivity Properties of Mainline BitTorrent DHT Nodes Raul - - PowerPoint PPT Presentation

connectivity properties of mainline bittorrent dht nodes
SMART_READER_LITE
LIVE PREVIEW

Connectivity Properties of Mainline BitTorrent DHT Nodes Raul - - PowerPoint PPT Presentation

Connectivity Properties of Mainline BitTorrent DHT Nodes Raul Jimenez, Flutra Osmani, Bjrn Knutsson KTH, Sweden 1 IEEE P2P'09 9-11 Sept Seattle, WA Background P2P-Next Large EU project Content distribution platform on


slide-1
SLIDE 1

1

Connectivity Properties of Mainline BitTorrent DHT Nodes

Raul Jimenez, Flutra Osmani, Björn Knutsson KTH, Sweden

IEEE P2P'09 – 9-11 Sept – Seattle, WA

slide-2
SLIDE 2

2

Background

  • P2P-Next

– Large EU project – Content distribution platform on top of BitTorrent

  • Kademlia DHT (Mainline implementation)

– Used as distributed BitTorrent tracker – Iterative routing algorithm

slide-3
SLIDE 3

3

Kademlia: lookup

slide-4
SLIDE 4

4

Kademlia: lookup

20% dead nodes in 95% of the lookups One minute lookups [Crosby-07] Non-transitive connectivity and DHTs [Freedman-05]

slide-5
SLIDE 5

5

Connectivity Properties

Transitivity Persistence

B C A t0 t1 A B

Reciprocity

B A

slide-6
SLIDE 6

6

Experiment

  • Nodes' connectivity properties
  • 3,6 million unique nodes in 24 hours

R D N L Same IP N triggers experiment Up to 5 queries per vantage point 1 minute timeout Repeat after 5 minutes

slide-7
SLIDE 7

7

Experiment

Reciprocal Connectivity

D N D N

Reciprocity

N D

slide-8
SLIDE 8

8

Experiment

Reciprocal Connectivity

D N

80 %

D N

20 %

Reciprocity

N D

slide-9
SLIDE 9

9

Experiment

Transitive Connectivity

R D N

Transitivity

D N R

slide-10
SLIDE 10

10

Experiment

Transitive Connectivity

R D N

40 %

Transitivity

D N R

slide-11
SLIDE 11

11

Persistent Connectivity

Experiment

D N t t + 5 minutes

Persistence

t0 t1 D N

slide-12
SLIDE 12

12

Persistent Connectivity

Experiment

D N

44 %

t t + 5 minutes

Persistence

t0 t1 D N

slide-13
SLIDE 13

13

Experiment Results

  • Clear overlap: transitivity & persistence
  • Just 1/3 of the nodes show “good” connectivity
slide-14
SLIDE 14

14

Discussion

  • Possible causes

– Firewalls

Reciprocity →

– NATs

Transitivity and/or persistence →

– DHT implementation

All →

  • Potential effects on the DHT performance

– Higher (apparent) churn – Broken lookup routes

slide-15
SLIDE 15

15

Conclusion

  • Understand the underlying connectivity issues

– Explicit definition of connectivity properties – Large scale experiment on a real-world DHT

  • Use this knowledge to achieve faster lookups
slide-16
SLIDE 16

16

Thank you!

slide-17
SLIDE 17

17

t t + 5 min % Obs

10.6

Firewall

31.3

Port restricted cone & Symmetric NAT

2.7

Full cone NAT & Real churn

35.5

Open Internet

Possible Causes

R D N R D N R D N R D N R D N R D N R D N R D N

slide-18
SLIDE 18

18

Experiments

slide-19
SLIDE 19

19

NAT types

  • Match IP and port

(port restricted cone & symmetric)

IP2:2222 IP1:1111 IP3:3333 IP1:9999

slide-20
SLIDE 20

20

NAT types

  • Match only IP

(restricted cone)

IP2:2222 IP1:1111 IP3:3333 IP1:9999

slide-21
SLIDE 21

21

NAT types

  • Match any connection

(full cone)

IP2:2222 IP1:1111 IP3:3333 IP1:9999 Reference: RFC3489 STUN - Simple Traversal of User Datagram Protocol (UDP) Through Network Address Translators (NATs)

slide-22
SLIDE 22

22

What to do?

  • Prevent my routing table from pollution

– Check for global reachability before adding a

node

  • Don't pollute others' routing tables

– When I'm aware of being partially reachable, tell

  • thers: “don't add me to your routing table”
  • Free riders?