An Experimental Study of the Skype PeertoPeer VoIP System

1
An Experimental Study of the Skype Peer-to-Peer
VoIP System

• Saikat Guha, Cornell University
• Neil DasWani, Google
• Ravi Jain, Google
• IPTPS06
• Presenter Te-Yuan

2
What do they want to know?

• What makes Skype so successful?
• Compare with
• File-sharing P2P network
• By Observing Skypes
• User behavior
• Node Session Time
• Overlay Network Traffic
• SuperNode overlay network
• Overall utilization resource consumption

3
Skype

• Three Services
• two-way audio streams conference call up to 4
  users
• Instant Message
• file-transfer
• Structure
• Alike KaZaA SuperNode-based
• Ordinary Node (ON)
• Super Node (SN)

4
Outline - Experiments

• Expt. 1 Basic operation
• Expt. 2 Promotion to supernode
• Expt. 3 Supernode network activity
• Expt. 4 Supernode and client population
• Expt. 5 Supernode presence

5
Expt. 1 Basic operation

• To Answer How do two Skype clients connect to
  each other?
• Normally,
• ON send control traffic through SN-p2p
• Including
• Availability information
• Instant messages
• Request for VoIP File-transfer
• What if ON is behind NAT/Firewall?

6
Expt. 1 Basic operation Cont.

• NAT Traversal in Skype
• Level 0 Initiator NATed
• Level 1 Recipient NAT'ed
• Level 2 Both NAT'ed (well-behaved NATs)
• Level 3 Both NAT'ed

7
Expt. 1 Basic operation Cont.

• Level 0 Initiator NATed

8
Expt. 1 Basic operation Cont.

• Level 1 Recipient NATed

9
Expt. 1 Basic operation Cont.

• Level 2 Both NAT'ed (well-behaved NATs)

10
Expt. 1 Basic operation Cont.

• Level 3 Both NAT'ed

11
Expt. 1 Basic operation Cont.
Level 0
Level 1
Level 2
Level 3
12
Outline - Experiments

• Expt. 1 Basic operation
• Expt. 2 Promotion to supernode
• Expt. 3 Supernode network activity
• Expt. 4 Supernode and client population
• Expt. 5 Supernode presence

13
Expt. 2 Promotion to supernode

• To Answer What kind of node will be promote to
  SN?
• Setup several Skype clients
• One behind a saturated network uplink
• One behind a NAT
• One with a 10 Mbps connection public IP
• Key to be SN
• plenty of spare bandwidth
• publicly reachable

14
Outline - Experiments

• Expt. 1 Basic operation
• Expt. 2 Promotion to supernode
• Expt. 3 Supernode network activity
• Expt. 4 Supernode and client population
• Expt. 5 Supernode presence

15
Expt. 3 Supernode network activity

• Goal To observe the network traffic of a Skype
  supernode
• Duration 135 days (Sep. 1, 2005 to Jan. 14,
  2006)
• Data captured 13GB with ethereal

16
Expt. 4 Supernode and client population

• Goal Collect SN client IP/port
• Duration2005/7/25 2005/10/12
• Result
• Crawl 150K SN
• Collect 250K SN info

17
Expt. 4 Supernode and client population
Connect to a SN
A list of SN
Save the list
Connect to a SN from the list
A list of SN
18
Expt. 4 Supernode and client population

• Collect client info
• Collect the number reported by skype client

19
Expt. 5 Supernode presence

• Goal how many SN online at a give time
• Flow
• Randomly Select 6000 SN - from the list
  collected by expt. 4
• Send application-layer Ping
• Repeat every 30 mins for a month

20
Expt. 5 Supernode presence - Cont
Num. of SuperNode is more Stable
diurnal behavior of SN
Weekend
21
Expt. 5 Supernode presence - Cont

• Geographic Distribution of Active SuperNodes

15-25
20-25
45--60
peak at 11am UTC (Europe mid-day)
22
Expt. 5 Supernode presence - Cont

• SuperNode Session Time

Median is 5.5h
23
Expt. 5 Supernode presence - Cont

• Fraction of supernodes joining or departing

Node arrival concentrated toward morning
Skype usage is correlated with working hours
Different from P2P file-sharing
Node departureconcentrated toward evening
24
Expt. 5 Supernode presence - Cont

• Node Arrival dependent on Time
• Not Poisson or Uniform process
• Poisson process with varying hourly rate

Node arrival concentrated toward morning
Node departureconcentrated toward evening
25
VoIP in Skype Preliminary Observation
90.4SN no need to relay VoIP traffic

• SuperNode Traffic

26
VoIP in Skype Preliminary Observation

• VoIP Relayed Session Arrival Behavior
• Inter arrival time of Relayed VoIP/File sessions
  may be Poisson

27
VoIP in Skype Preliminary Observation

• VoIP Session Length Behavior

Skype Median 2m50s Average 12m53s Longest 3h
26s
Traditional Average 3m
Fraudulent Average 9m
28
VoIP in Skype Preliminary Observation

• File-transfer sizes

File size Median 346kB
29
Conclusion

• First measurement study of Skype VoIP system
• Skype differs significantly from file-sharing P2P
• User Behavior
• Diurnal Work-week
• Calls are significantly longer
• File transferred are significantly smaller
• SuperNode of Skype
• Consume little bandwidth
• Relatively stable

30
My Opinion

• Pros