volker hilt [email protected] bell labs/alcatel-lucent sip overload control ietf design team...

15
Volker Hilt [email protected] Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Upload: meryl-dickerson

Post on 28-Dec-2015

225 views

Category:

Documents


3 download

TRANSCRIPT

Page 1: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Volker Hilt

[email protected]

Bell Labs/Alcatel-Lucent

SIP Overload ControlIETF Design Team Status

Page 2: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 2 | 74 IETF Meeting | March 2009

SIP Overload Control Design Team

Team Members Eric Noel, Carolyn Johnson (AT&T 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 AT&T Labs, Bell Labs/Alcatel-Lucent, Columbia University, Sonus

Networks

Bi-weekly conference calls.

Page 3: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 3 | 74 IETF Meeting | March 2009

draft-ietf-sipping-overload-design-01Changes to -00

Added new sections on: Fairness

Introduces fairness categories.

Performance Metrics Discusses metrics to compare overload control mechanisms

Message Priorization Selection of messages in overload condition.

Added text to Security Considerations section.

Minor edits throughout the text.

Page 4: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 4 | 74 IETF Meeting | March 2009

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

Discussion of overload control mechanisms needs to be structured along the identified performance metrics.

Document is close to completion.

Page 5: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 5 | 74 IETF Meeting | March 2009

SIP Overload Control Design TeamSimulation Results

Four types of overload control Rate-based Overload Control Loss-based Overload Control Window-based Overload Control Overload Signal-based Overload Control

Summary of Steady-State Evaluation (presented at IETF ’73)

Performance of all overload control mechanisms under evaluation is similar in steady state.

Varying network conditions (i.e., delay, loss-rate) do not reveal significant differences.

Results for Transient Scenarios

Evaluation of transient behavior with respect to Changes in offered load changes in the number of neighbors Fairness

Page 6: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 6 | 74 IETF Meeting | March 2009

Changes in Offered-Load (AT&T Labs) Rate-based and Window-based Overload Control

Simulations use the following overload control feedback types and algorithms:

Rate-based: queue delay Loss-based: SRED Window-based

Feedback conveyed in SIP responses.

Result: rate-, loss- and window-based controls respond well to transient stimulus.

60

80

100

120

140

160

180

0.00 200.00 400.00 600.00 800.00Time (sec)

Car

ried

load

(cp

s)

TheoreticalQueue delay control

60.00

80.00

100.00

120.00

140.00

160.00

180.00

0.00 200.00 400.00 600.00 800.00Time (sec)

Car

ried

load

(cp

s)

TheoreticalWindow contorl

60

80

100

120

140

160

180

0 200 400 600 800Time (sec)

Car

ried

load

(cp

s)

TheoreticalSRED control (fmin=0.03)

Page 7: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 7 | 74 IETF Meeting | March 2009

Changes in Offered-Load (Bell Labs/Alcatel-Lucent) Loss-based and Rate-based Overload Control

0

100

200

300

400

500

600

700

0 100 200 300 400 500 600

Load

Rate feedback

Loss feedback

Time

0

100

200

300

400

500

600

700

0 500 1000 1500 2000 2500 3000

Load

Rate feedback

Loss feedback

Overload control feedback type and algorithms used: SRED algorithm Loss- vs. rate-based feedback

SIP responses convey feedback from core to edge proxies.

Result: loss- and rate-based overload control perform well.

Time

CP

S

CP

S

Page 8: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 8 | 74 IETF Meeting | March 2009

Changes in Offered-Load (Columbia University) Window-based and Rate-based Overload Control

23/4/19 Slide 8

Window- and rate-based controls perform well.

Page 9: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 9 | 74 IETF Meeting | March 2009

Changes in Offered-Load (Sonus Networks) Loss-based and Overload-Signal-based Overload Control

Loss-based Overload Control

0

200

400

600

800

1000

1200

1400

1600

0 300 600 900 1200 1500 1800 2100 2400 2700

Time (Secs)

Off

ered

lo

ad (

cps)

Loss-based Overload Control

60

80

100

120

140

160

180

0 300 600 900 1200 1500 1800 2100 2400 2700

Time (Secs)

Car

ried

lo

ad (

cps)

Target Overload Signal Rate =10 Overload Signals /Sec

Overload Signal-based

60

80

100

120

140

160

180

0 300 600 900 1200 1500 1800 2100 2400 2700

Time (Secs)

Car

ried

lo

ad (

cps)

Overload Signal-based

0

200

400

600

800

1000

1200

1400

1600

0 300 600 900 1200 1500 1800 2100 2400 2700

Time (Secs)

Off

ered

lo

ad (

cps)

Page 10: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 10 | 74 IETF Meeting | March 2009

Changes in the Number of Senders (Bell Labs/Alcatel-Lucent)Loss-based and Rate-based Overload Control

Edge proxies are turned on/off sequentially.

Each edge proxy sends the same amount of load while active.

Feedback-type and algorithms: Rate-fixed: core proxies are configured with a fixed number of senders.

The overall rate of a core proxy is divided through the sender number. Rate-aware: core proxies estimate the number of senders. The overall

rate of a core proxy is divided through the sender estimate.

Loss-based: same loss rate is sent to all edge proxies. All simulations use SRED algorithm.

Page 11: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 11 | 74 IETF Meeting | March 2009

Fairness (Columbia University)Rate-based Overload Control

Slide 11

Provider-centric fairness: each source gets the same share

User-centric fairness: each source gets a share proportional to its original incoming load

Page 12: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 12 | 74 IETF Meeting | March 2009

Conclusion & Next Steps

Simulation Results

The overload control performance seems to differ little between the type of feedback:

Rate-, Loss-, Window- and Signal-based mechanisms all performed well in steady-state as well as transient evaluations.

Of course, the performance does vary depending on the overload control algorithms used and parameter settings of these algorithms.

But: algorithms and parameter settings are likely to be out of scope for an overload control protocol specification.

Next Steps

Evaluate additional transient scenarios.

Finalize draft-ietf-sipping-overload-design-01

Work on a solution!!

Page 13: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Session Initiation Protocol (SIP) Overload

Controldraft-hilt-sipping-overload-06

Volker Hilt, Indra Widjaja, Henning Schulzrinne

A Session Initiation Protocol (SIP) Load Control Event

Packagedraft-shen-sipping-load-control-event-package-01

Charles Shen, Henning Schulzrinne, Arata Koike

Page 14: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 14 | 74 IETF Meeting | March 2009 14

Service Provider B

Hotline Callee212-555-12349am-10am,

2009-1-1Service Provider A

Enterprise Network B

Enterprise Network A

Filter SpecID: To: +1-212-555-1234Time: 9am-10am 2009-1-1Act: accept rate= Nmax

Filter SpecID: To: +1-212-555-1234Time: 9am-10am 2009-1-1Act: accept rate=NSPA

Filter SpecID: To: +1-212-555-1234Time: 9am-10am 2009-1-1Act: accept rate=NEPA

Filter SpecID: To: +1-212-555-1234Time: 9am-10am 2009-1-1Act: accept rate=NSPB

Charles Shen, Henning Schulzrinne, Arata Koike, A Session Initiation Protocol (SIP) Load Control Event Package, draft-shen-sipping-load-control-event-package-01.txt, IETF SIPPING Working Group, Work in Progress. Nov 3, 2008

Filter-based SIP Server Overload Controldraft-shen-sipping-load-control-event-package-01

Page 15: Volker Hilt volkerh@bell-labs.com Bell Labs/Alcatel-Lucent SIP Overload Control IETF Design Team Status

Slide 15 | 74 IETF Meeting | March 2009

Server S1 Server S2

Session Initiation Protocol (SIP) Overload Controldraft-hilt-sipping-overload-06

Overload control mechanism Enables proxies to send overload control feedback to upstream

neighbors. Feedback is conveyed in SIP responses

New Via Header Parameters

Supports different types of feedback. Currently defined: loss-based. Specifies the protocol semantics. Open to different overload control

algorithms.

Via: SIP/2.0/TCP ss1.example.com:5060 ;oc=20;oc_validity=500

OverloadReduces load