A simulation-based comparative evaluation of transport protocols for SIP

1
A simulation-based comparative evaluation of
transport protocols for SIP

• Authors M.Lulling, J.Vaughan
• Department of Computer science,
• University college Cork,
• Western Road, Cork, Ireland.
• Publication ELSEVIER on Computer
• communications, April 2005
• Reporter Chun-Hui Sung
• Date 2007/5/24

2
Outline

• Introduction
• Transport for SIP
• Simulations
• Results
• Conclusion
• Comment

3
Introduction

• Uses the Network Simulator NS2 to investigate
  the direct effects and subsequent consequences
  associated with the use of different transport
  protocols in a SIP context .
• Performance evaluation in the result of VoIP SIP
  signaling from simulation-based experiments
  underlying transport protocol.

4
Introduction ( Cont. )

• SIP (Session Initiation Protocol) is
• A peer-to-peer protocol
• An application layer signaling protocol
• Create, modify and terminate sessions
• Applications can be voice, video, gaming, instant
  messaging, presence, call control, etc.

5
Transport for SIP

• SIP over TCP
• TCP Reno
• TCP Vegas
• TCP Sack
• SIP over UDP
• SIP over SCTP

6
SIP over TCP

• TCP - Reno

7
SIP over TCP

• TCP - Vegas

8
SIP over TCP

• TCP - Sack

9
SIP over UDP

• The user datagram protocol, UDP, is a
  connectionless transport protocol that does not
  provide any guarantee of message delivery.

10
SIP over SCTP

• The stream control transmission protocol, SCTP,
  is a reliable end-to-end transport layer
  protocol, and while support for TCP and UDP is
  included in the core SIP specifiation.

11
Simulations

• Network topology
• Node 1 and 2 are buffer-limited droptail routers,
  all other nodes are endpoints, node 1 and 2 are
  the only bottleneck link.
• Node 0 and 3 are SIP proxies.
• Node 4 and 5 are used to provide competing
  cross-traffic.

12
Simulations ( Cont. )

• NS2 parameters
• Delay time are 45 ms between the proxies.
• The simulations use a stationary Poisson model
  to generate the arrival times of 512-byte session
  establishment requests at node 0.
• Individual SIP requests are independent and are
  generated at node 0 at 160/s, which corresponds
  to a link utilization of approximately 33 on the
  bottleneck link.

13
Simulations ( Cont. )

• Induced packet loss
• Random packet loss
• Competing traffic
• Throughput analysis

14
Simulations ( Cont. )

• Induced packet loss
• In order to measure and evaluate the delays and
  delaying effects of packet loss on the system,
  packets are explicitly dropped from node 1.
• The simulation is run 10 times for each of the
  five transport protocols or variants.
• The time at which each message is generated by
  the application at node 0 and the time at which
  this message is passed to the application at node
  3 is recorded. (delay time)

15
Simulations ( Cont. )

• Random packet loss
• Random packet loss percentages of between 0.1 and
  0.5 (in 0.1 intervals) are simulated at node 1
  with uniform distribution.
• The time at which each message is generated by
  the application at node 0 and the time at which
  this message is passed to the application at node
  3 is recorded. (delay time)

16
Simulations ( Cont. )

• Competing traffic
• Simulate the effects of cross-traffic generated
  between node 4 and 5, providing competition for
  bandwidth on the bottleneck link between nodes 1
  and 2.
• TCP Reno is used exclusively as the transport
  protocol for the competing traffic in all
  simulations.
• Delays are measured as describe in the two
  previous experiments.

17
Simulations ( Cont. )

• Throughput analysis
• Add a variable of buffer size at node 1
• The simulations have been run with buffer sizes
  of 5, 20, 50, 100, 150, 200 and 250 packets at
  node 1.
• The default value of buffer size is 50.

18
Results

• Induced packet loss
• Random packet loss
• Competing traffic
• Throughputs

19
Results Induced packet loss (1/3) Five
consecutively dropped packets
20
Results Induced packet loss (2/3) Peak
delays
21
Results Induced packet loss (3/3) Message
affects
22
Results Random packet loss (1/4) loss rate of
0.3

• SIP traffic with TCP Reno

• SIP traffic with SCTP

23
Results Random packet loss (2/4) loss rate of
0.3

• SIP traffic with TCP Sack

• SIP traffic with UDP

24
Results Random packet loss (3/4) loss rate of
0.3

• SIP traffic with TCP Vegas

25
Results Random packet loss (4/4) loss rate
of 0.1 0.5

• Mean delay per packet loss percentage

26
Results competing traffic (1/4) SIP traffic
with TCP Reno
27
Results competing traffic (2/4) SIP traffic
with UDP / SCTP / TCP SACK
28
Results competing traffic (3/4) overall
throughput
29
Results competing traffic (4/4) Total
throughput
30
Results Throughputs

• Mean throughput for SIP vs. FTP TCP Sack

• Mean throughput for SIP vs. FTP TCP Vegas

31
Conclusion

• Authors compare and analyze the performance of
  SIP over UDP TCP-Reno/Vegas/Sack, SCTP.
• This paper was found that TCP Sack and SCTP are
  the best options for a reliable transport
  protocol for SIP traffic.

32
Comment

• They dont put attention on multi-homing of SCTP.