Transcript Measuring P2P IPTV Systems

Measuring P2P IPTV Systems
Thomas Silverston, Olivier Fourmaux
Universit ´e Pierre et Marie Curie - Paris 6
ACM NOSSDAV 2007
17th International workshop on Network and
Operating Systems Support for Digital Audio & Video
1
Outlines



Introduction
Experiment Setup
Measurement Analysis






General Observation
Traffic Pattern
Video Download Policy
Peers Neighborhood
Video Peers Lifetime
Conclusion
2
Introduction


P2P : ~70% overall Internet traffic
P2P applications


File sharing : BitTorrent, Kazaa, eDonkey, etc.
P2P streaming



P2P measurement studies

File sharing: BitTorrent, Kazaa, eDonkey


[Bharambe Infocom06], [Legout IMC06], [Liang Comp. Net.06]
VOIP: Skype, Google talk


IPTV(Live streaming) : PPStream, PPLive, CoolStreaming, etc.
Video on Demand(VoD) : Youtube, MSN Video, Dailymotion, etc.
[Baset Infocom06], [Suh Infocom06], [Barbosa Nossdav07], [Bonfiglio
Sigcomm07]
No comprehensive study about P2P streaming

Lots of academic P2P streaming protocols not really deployed on the
Internet


Anysee[Infocom06], Chunkspread[ICNP06], Prime[Infocom07], etc.
Commercial P2P streaming really deployed on the Internet



PPLive, PPStream, SOPCast and TVAnts
Proprietary softwate
No design/implementation information, patented.
3
Introduction(Cont.)

Need for P2P streaming measurements


P2P IPTV: massively used in the future
How do P2P streaming applications really works?




P2P video live streaming applications  P2P IPTV


PPLive, PPStream, SOPCast and TVAnts
Features



Data are divided into chunks
Each peer exchanges with other peers information about the chunks
Making comparisons between 4 different applications



Traffic analysis is the only feasible to identify the mechanisms
Link between academic and commercial
Input for model(simulation)
Highlighting design similarities and differences
Point out global behavior
Packet Traces

Two soccer games in 2006 FIFA World Cup on June 30, 2006



Large-scale event
Live interest for users
Real conditions
4
Experiment Setup

Two soccer games were scheduled on June 30, 2006

They are well representative of all of them


Four packet traces



With different applications at the same time
The first game(Germany vs. Argentine, in the afternoon) : PPStream, SOPCast
The second game(Italy vs. Ukraine, in the evening) : PPLive, TVAnts
Measurement experiment platform



Common PCs with 1.8GHz CPU
100Mbps Ethernet access (campus network environment)
tcpdump for Unix, ethereal for Windows XP http://www.ethereal.com/
5
Measurement Analysis

Packet traces summary
6
Measurement Analysis(Cont.)

Packet traces summary
TVAnts
is more
between
TCP and
UDP.
Major
part
of UDP.
PPLive
traffic
relies
TCP.
SOPCast
trafficbalanced
relies
mostly
PPStream
on
relies
only
onon
TCP.
7
Measurement Analysis
TVants
Fluctuating largely
Quiet constant
Total download and upload throughput for TVAnts.
8
Measurement Analysis
PPLive
9
Measurement Analysis
PPStream
10
Measurement Analysis
SOPcast
Received no traffic, but
PPStream was working well.
11
Traffic Pattern

Application features




Session duration



Video sessions are likely to have long duration
Signaling sessions are likely to be shorter in time
Packet size



Exchanging information about data chunks and neighbor peers
(Swarming mechanism)
Discovering other peers iteratively
Establishing new signaling or video sessions
Video streaming packet size is expected to be large
Signaling session packet size is suppose to be common
Average packet size according to peers session duration.
12
Traffic Pattern
Video
sessions
Signaling
sessions
Signaling
sessions
Signaling
sessions
They are not
clearly formed.
Video
sessions
A balanced use
of TCP and UDP
13
Traffic Pattern

Observations summary for traffic patterns

Signaling overhead

>= 1000 Bytes
Separating video and signaling traffic with an heuristic [6]

If a session had at least 10 large packets, then it was labeled as a
video session
Same IP addresses and ports
14
[6] X. Hei, C. Liang, J. Liang, Y. Liu, and K. W. Ross, “Insights into pplive: A measurement study of a large-scale p2p iptv
system,” in Proc. of IPTV Workshop, 2006.
Video Download Policy(VDP)
A major part of the
download traffic
Do not contribute to
a large part of the
download traffic
Almost all the
traffic during its
session duration
Neither the
top peer
About half the total
download traffic
(like SOPCast)
The problem did not occur
for network reasons
About half the total
download traffic
All the top ten
peers traffic during
its session duration
Not a large amount
of the total traffic
(like PPStream)
15
Total traffic, top ten peers traffic and top peer traffic
Video Download Policy(VDP)
Session duration

Short
PPLive



PPLive, SOPcast
PPStream
Getting the video from only a few peers at the same time
Switching periodically from a peer to another one
Peers
capacities
PPStream



Long
Low the data from many peers at the same time Huge
Getting
TVAnts duration
PPLive
Its peersPPStream,
have long session
SOPCast

Downloadofpolicy
like same
PPLive time
policy
The
number
VDPlooks
at the

Need
A fewmore than a peer to get the video compare to PPLive
Many
TVAnts
PPLive SOPCast TVAnts PPStream

Mix PPStream and SOPCast policies

Summary



The presented applications implement different download policies
Do not expect peers to have the same capabilities
16
Peers Neighborhood
Using an important
part of UDP traffic
High and constant
High and fluctuates
High and
fluctuates largely
Low and constant
17
Video Peers Lifetime

The video peer lifetime


The duration between the first time and the last time our
controlled nodes exchanging video traffic with another
peer.
End-hosts, similar to the tracker in BT, are
responsible to duplicate flows to each other



End-hosts can join and leave the network whenever they
want and are prone to suffer failures.
The systems have to deal with the arrivals and departures
of peers (churn of peer).
A high churn of peers will involve additional delays or jitter
variations for packet delivery, which will decrease overall
video quality.
18
Video Peers Lifetime

Video peers lifetime for TVAnts



All the applications have the same Weibull-like distribution for peers lifetime
The video peers lifetime CCDF follows a Weibull distribution
Complementary Cumulative Distribution Function (CCDF)
For all applications, there are no more
than 10 % of peers , which stay in the
network during an entire game (5400s).
TVants. Average lifetime = 2778s
5000s
19
Video Peers Lifetime
For all the applications, no more than 10% of peers stay in the
network during the entire game(5400s=1.5hr).
20
Video Peers Lifetime
21
Conclusion



We explored the behavior of 4 popular P2P IPTV systems by
measuring and analyzing their network traffic
Our analyses show that the measured applications generate
different traffic patterns and use different mechanisms to get
the video
This knowledge will be used in our other works to model and
simulate these systems
22
Reference

[6] X. Hei, C. Liang, J. Liang, Y. Liu, and K. W. Ross, “Insights into
pplive: A measurement study of a large-scale p2p iptv system,” in
Proc. of IPTV Workshop, 2006.





X. Hei, C. Liang, J. Liang, Y. Liu, K.W. Ross, “A Measurement Study of a
Large-Scale P2P IPTV System,” IEEE Transactions on Multimedia,vVol.9,
No.8, pp.1672-1687, Dec. 2007.(Journal version)
[7] K. Sripanidkulchai, A. Ganjam, B. Maggs, and H. Zhang, “The
feasibility of supporting large-scale live streaming applications with
dynamic application end-points,” in Proc. of SIGCOM, 2004.
[8] X. Zhang, J. Liu, and B. Li, “On large-scale peer-to-peer live video
distribution: Coolstreaming and its preliminary experimental results,”
in Proc. MMSP, 2005.
[10] T. Silverston and O. Fourmaux, “P2p iptv measurement: A
comparison study,” http://www.arxiv.org/abs/cs.NI/0610133, 2006.
Eugenio Alessandria, Massimo Gallo, Emilio Leonardi, Marco Mellia,
Michela Meo, “P2P-TV Systems under Adverse Network Conditions:
a Measurement Study,” IEEE INFOCOM 2009.
23