Volker%20Hilt

1
SIP Overload ControlIETF Design Team Status

• Volker Hilt
• volkerh_at_bell-labs.com
• Bell Labs/Alcatel-Lucent

2
SIP Overload Control Design TeamCurrent Status

• Team Members
• Eric Noel, Carolyn Johnson (ATT Labs)
• Volker Hilt, Fangzhe Chang (Bell
  Labs/Alcatel-Lucent)
• Charles Shen, Henning Schulzrinne (Columbia
  University)
• Ahmed Abdelal, Tom Phelan (Sonus Networks)
• Mary Barnes (Nortel)
• Jonathan Rosenberg (Cisco)
• Nick Stewart (British Telecom)
• Four independent simulation tools
• ATT Labs, Bell Labs/Alcatel-Lucent, Columbia
  University, Sonus Networks
• Bi-weekly conference calls.

3
draft-ietf-sipping-overload-design-00Status and
Changes

• Status
• Submitted as SIPPING WG item.
• Changes to draft-hilt-sipping-overload-design-00
• Numerous clarifications throughout the text
• Added separate section for problem description
• Added an overload control mechanism
• Overload signal-based overload control
• Discussion of implicit overload control in
  separate section.

4
draft-ietf-sipping-overload-design-00Next Steps

• Document investigates design choices and models.
• Document is stable.
• Simulation results are not included.
• Ready for WGLC?

5
Simulation Results Overload Control Feedback
Types

• Rate-based Overload Control
• Limit the request rate a server receives from an
  upstream element.
• Feedback X requests per second.
• Loss-based Overload Control
• Reduce the request rate a server receives by a
  percentage X.
• Feedback reduce load by X.
• Window-based Overload Control
• Limit the number of request a server can receive
  without confirming a request.
• Feedback send X more requests.
• Overload Signal-based Overload Control
• De-/increase of request rate until a target
  overload notification rate is reached.
• Feedback server overloaded (503 response without
  RetryAfter header)

6
Simulation ResultsNo Overload Control

• Simulation Setup
• Evaluate SIP overload performance under varying
  network conditions.
• Network delay 0 2 sec
• Network loss 0 50
• Topology consists of 5 edge proxies and 2 core
  proxies.
• Overall capacity 142 cps
• Three levels of offered load
• Underload 114 cps
• Overload 500 cps, 1000 cps
• Results
• Delay and loss substantially decrease goodput
  even when there is no overload.
• Confirmed by all simulators (ATT Labs, Bell
  Labs/Alcatel-Lucent, Columbia University, Sonus
  Networks)

7
Simulation ResultsUpper Bound Estimation

• Delay and loss triggers retransmissions.
• Increases the number of SIP messages needed to
  set up a call.
• Causes the goodput of a server drops.
• Upper bound estimation
• Determine increase in message count caused by
  delay/loss.
• Example topology UAC proxy - UAS
• Calculate resulting goodput.
• Example result 125 ms delay
• Round trip delay exceeds 500ms.
• One 200 OK/ACK retransmission and one BYE/200 OK
  retransmission triggered.
• Messages per call increases from 7 to 11.
• Contributor ATT Labs

8
Simulation ResultsSelected Results on Overload
Control (1)
Rate-based Control
Loss-based Control
ATT Labs
Bell-Labs/Alcatel-Lucent
9
Simulation Results Selected Results on Overload
Control (2)
Overload Signal-based Control
Window-based Control
Sonus Networks
Columbia University
10
Conclusions and Next Steps

• Conclusions
• Large delay and loss rates can cause the goodput
  to drop substantially even in non-overloaded
  conditions.
• Performance of all overload control mechanisms
  under evaluation is similar in steady state.
• Varying network conditions do not reveal
  significant differences.
• Next Steps
• Evaluate additional scenarios.
• Impact of different load distributions
• Impact of varying upstream neighbor counts
• Transient simulations.
• Evaluate dynamic behavior of overload control
  mechanisms
• Transition in/out of overload, impact of load
  peaks, etc.
• Evaluate fairness of overload control mechanisms.

11
SIP Overload Control Design TeamPublications

• E. Noel, C. Johnson, Initial Simulation Results
  That Analyze SIP Based VoIP Networks Under
  Overload, International Teletraffic Congress
  (ITC07), Ottawa, Canada, June 2007.
• C. Shen, H. Schulzrinne, E. Nahum, Session
  Initiation Protocol (SIP) Server Overload
  Control Design and Evaluation, Principles,
  Systems and Applications of IP Telecommunications
  (IPTComm08), Heidelberg, Germany, July 2008.
• V. Hilt, I. Widjaja, Controlling Overload in
  Networks of SIP Servers, IEEE International
  Conference on Network Protocols (ICNP08),
  Orlando, Florida, October 2008.