Comparing BitTorrent Clients in the Wild:

The Case of Download Speed

Georgos Siganos, Xiaoyuan Yang, and

Marios Iliofotou
Pablo Rodriguez
University of California, Riverside Telefonica Research, Barcelona

April 27th ! IPTPS 2010 ! San Jose, CA

TELEFÓNICA I+D

© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal

Motivation

!! BitTorrent (BT) is the most popular P2P file-sharing protocol

—! 20 million daily users in Pirate Bay alone

!! Today, there exist many BT clients

—! How much do we know about them?

uTorrent
etc.
Vuze Transmission

Question: Are there differences between these clients?

TELEFÓNICA I+D
1
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Contributions

!! Large scale measurement study of real BT clients

—! 10 million users over one month
—! From 6,000 ISP

!! Observation 1: uTorrent achieves on average 16% higher

download speeds compared to Vuze

!! Observation 2: We indentify four implementation differences

—! Neighborhood management
—! Opening of new connections
—! Termination of connections
—! Upload bandwidth distribution

TELEFÓNICA I+D
2
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Outline

!! Introduction

!! Dataset collection

!! Part A: Speed comparison

!! Part B: Implementation differences

!! Conclusions and future work

TELEFÓNICA I+D
3
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Download speed inference using Apollo

BT host
BT host

Get ~ 7.5 minutes ~

bitfield Get
bitfield
Apollo
Apollo

Time
1 0 1 0 1 0 0 0 1 0 1 0 1 1 1 1

1 0 1 0 1 1 1 1
1 0 1 0 1 0 0 0 ! # Pieces " Size
Estimated ! Speed =
Time
Difference = 3 pieces

TELEFÓNICA I+D
4
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Dataset description

!! Collection process
—! Apollo connects to the top 600 torrents from Pirate Bay (every hour)
—! It connects multiple times with as many BT hosts possible

!! BT Clients used: Vuze and uTorrent

—! 75% of all BT hosts use these clients (Pirate Bay)

!! Summary statistics
—! Data cover 1 month (a representative week is used in the paper)
—! 10 millions BT hosts
—! 6,000 ISPs

TELEFÓNICA I+D
5
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Fair comparison between Vuze and uTorrent

!! External factors that can effect download speed

—! The torrent: A local torrent can be faster (e.g., a Spanish film)
—! The ISP: An ISP can be faster than others

!! Solution: Select same # hosts from each {torrent, ASN} pair

!! Data after processing

—! Overall equal number of Vuze and uTorrent samples
—! 10K up to 80K samples per ASN for each client

TELEFÓNICA I+D
6
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Outline

!! Introduction

!! Dataset collection

!! Part A: Speed comparison

!! Part B: Implementation differences

!! Conclusions and future work

TELEFÓNICA I+D
7
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Overall speed differences
200

Average Download Speed (Kbps)

180
Perc. uT Vuze Ratio 160 16%
(Kbps) (Kbps)
30th 65 61 1.07 140
50th 90 81 1.11 120
60th 119 106 1.12 100
70th 165 147 1.13
80
Two main observations: 60
•! uTorrent faster by 16% 40
•! uTorrent is faster for
the higher percentiles 20
0
uTorrent Vuze

TELEFÓNICA I+D
8
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Consistent behavior over time
Average download speed for the two clients

TELEFÓNICA I+D
9
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Comparing speeds for different ISPs

Some ISPs 20% of ISPs

very similar > than 25%
speeds

TELEFÓNICA I+D
10
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Outline

!! Introduction
!! Dataset collection
!! Part A: Speed comparison
!! Part B: Implementation differences
—! Neighborhood management
—! Opening of new connections
—! Termination of connections
—! Upload bandwidth distribution

!! Conclusions and future work

TELEFÓNICA I+D
11
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Data collection

!! Trace driven reconstruction of internal state

—! Keep track of active connections
—! Estimate upload and download rate of each connection
!! Why this approach?
—! It can be applied to any client (open or close source)
—! We have the same measurement methodology (no biases)
!! Dataset
—! Experiments ranged over two months
—! Different times of the days, weekdays and weekends
—! Always downloading a popular torrent (top 20 in Pirate Bay)

TELEFÓNICA I+D
12
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Neighborhood set over time
uTorrent Vuze
50
50
35
29
15
13
10

!! Both clients are set to have 50 open connections

—! Vuze follows this closely
—! uTorrent shows higher variability
!! Vuze has more ephemeral connections (open for less than 5 mins)

TELEFÓNICA I+D
13
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Upload bandwidth distribution

Vuze: uT:
30% 90%

!! uTorrent simultaneously uploads to more peers

—! uTorrent 90% of the time uploads to more than 4
—! Vuze only 30% of the time uploads to more than 4

TELEFÓNICA I+D
14
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Outline

!! Introduction

!! Dataset collection

!! Part A: Speed comparison

!! Part B: Implementation differences

!! Conclusions and future work

TELEFÓNICA I+D
15
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Conclusions

!! BT is popular, but how much do really we know about it?

—! Our measurements open many interesting questions

!! Our goal is to bring to the attention of BT designers that

—! Some design choices can have a significant effect in performance
—! More effort should be added in evaluating design choices

!! We don’t claim that uTorrent is the way to go

—! We see our work opening the door for new research efforts
to measure and evaluate real BT clients in the wild

TELEFÓNICA I+D
16
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
Future work

!! Compare more BT clients

—! E.g., transmission, bitcommet, etc.

!! Test different implementation choices in a

controlled environment
—! Testbed
—! Planetlab

!! Study the effect of the ISP and their offered

services to the performance of different clients

TELEFÓNICA I+D
17
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal
 Thank You!
Questions/Discussion?

TELEFÓNICA I+D
18
© 2010 Telefónica Investigación y Desarrollo, S.A. Unipersonal