1

OpMiGua, a Circuit/packet hybrid optical network enabling absolute service guarantees and high throughput

Steinar Bjørnstad Norwegian University of Science and

Technology (NTNU) Institute of Telematics

OpMiGua homepage: www.opmigua.com

2

Challenge: Migration from circuitto packet networks• Candidate: Optical packet/circuit hybrid

networks that brings together the best from two worlds– QoS and cost efficiency

3

Outline• Background• Motivation• QoS demands in broadcast and mobile networks• Introduction to the OpMiGua hybrid principle• Node and network architecture• Performance simulations• Experimental results• Further work• Summary

4

OpMiGua

• Optical Packet Switched Migration capable Network with Guaranteed service– Project proposed by S. Bjørnstad 2003, ended 2006

• Explore the packet/circuit hybrid network• Funded by Norwegian research council• Participants

– Telenor, Norwegian Univ. Science and Technology, Network Electronics

• Homepage: www.opmigua.com

5

Motivation• Network capacity needs increases• Packet based transport replaces circuit based

transport– Not straight forward: Added latency, jitter– Potentially higher power consumption and cost.

• Desire for a converged network serving multiple services – Telephony, file transfer, IP-TV, TV-broadcast,….

• Converged network requirements– Efficiency of packet switching (statistical multiplexing)– QoS of circuit switching (high reliability and performance)

6

TV-broadcast network structure

Distributionnetwork

Operator distributionnetwork

Distributionnetwork

Head-End

Distributionnetwork

Head-End

Head-End

Broadcastercontributionnetwork

7

QoS demands in TV-broadcasting• Contribution

– Between production facilities– A single packet loss affects all viewers– Packet loss or any type of interruption should not occur

• Primary distribution– From broadcaster to TV-transmitters or IP-TV head-ends– Information loss affects all viewers in a segment

• Secondary distribution– From IP-TV head-end or TV-transmitter to the individual viewers. – Information loss typically affects a high number of viewers

8

Broadcast performance requirements• Viewers call customer support when artifacts occur in

the picture– Especially during films and sports events like e.g. Olympic games– Expensive!– PLR demand < 10-8

• Pictures must not be skipped or repeated at receiver side– Synchronization between transmitter and receiver needed– Difficult in asynchronous IP-networks, GPS needed

9

Mobile backhaul QoS requirements

• Migration from circuit (PDH) to packet (Ethernet)– Mobile data-traffic drives the demand

• Voice traffic demands low packet loss and low delay• Handover between base-stations demands predictive

delays and synchronization– Low or no delay jitter, alternatively GPS

10

The OpMiGua hybrid network• Packet based transport but circuit/packet switched• ORION (Ghent univ. and part of STOLAS proj.)

Related principle and project• Time multiplexed packet and circuit switched network

– Some packets follow wavelength paths (circuits) – Some packets are switched by packet switches

• Packets are tagged to decide: Wavelength path or packet switching– OpMiGua uses polarizations state as tag

11

GST lightpath from A→D

A DCB

D

D

D

D

Pure Circuit switched system

Incoming GST packets destined for node D

- Traffic is routed according to circuit e.g. wavelength → minimal delay ☺- Traffic to different destinations must use different circuits or wavelengths

– Ensures no collisions between the two traffic streams ☺

– Requires one wavelength per stream

–Streams may e.g. be one circuit and one packet: Classic hybrid

C

C

C

C

12

Pure packet switched system

Ingr

ess

queu

e

Ingr

ess

queu

e

Ingr

ess

queu

e

Egr

ess

queu

e

Egr

ess

queu

e

Egr

ess

queu

e

Transit queue Transit queueTransit queue

A B C

•Single circuit (may be several wavelengths) for transmission•All packets are processed electronically: Header address lookup, queuing, forwarding•Packets from Ingress queue and Transit queue are statistically multiplexed

13

Pure packet switched system

A B C

•Packets from A and B are destined to C•Different streams have different color•Interleaving of simultaneously arriving packets gives variable packet delays

C

C

C C

C

14

Pure packet switched system

A B C

•Packets from A and B are destined to C•Different streams have different color•Interleaving of simultaneously arriving packets gives variable packet delays

CCCCC

C

15

Pure packet switched system

A B C

•New arrivals at B destined for C•Transmit queue is full => packet drop in ingress node B !

CCCCC C

C

C

C

C

16

Pure packet switched system

•Packets arrive reordered at output •Packets may be dropped •Efficient utilization of transmission wavelength ☺

C

C

C

CC CC

C

C

C

A B C

17

DB

GST lightpath from A→D

Underutilized circuit (wavelength)

Incoming GST packets destined for node D

Transit traffic is not being processed in B and C

Time between packets is unused

– Pure WDM system (circuit) gives low channel utilization

– For 270 Mb/s video on a 1 Gbps, 70 % of capacity is wasted

OpMiGua switch will insert lower quality (Internet) traffic in voids

B

C

C

Packet Switch

D

Input Queue Output Queue

A C

D

CD

D

D

C

18

Two basic QoS classes

• Guaranteed Service Class (GST)– No logical packet loss, fixed delay, no packet jitter– Fixed delay allows synchronous transfer if desirable– Suitable for broadcast-TV and mobile backhaul

• Statistical Multiplexed (SM)– Statistical multiplexed packet switched– Suitable for less demanding services, e.g. file transfer

19

OpMiGua network nodes• Packets are detected at input SM or GST• SM packets are only scheduled if no GST packet is

detected at input. – A delay through the node ensures that preemption of packets at the

output does not occur.

DelayGST λ1

Sense SM add

Delay

SM drop/GST sense SM add

DelayGST λ2

SM drop

GST λ2

GST λ1

Ingress node Core node Egress node

Aggregation interfaces Aggregation interfaces

20

Node architecture• Optical cross connect• Optical or electronic packet switch• SM packet scheduling always finishes

– GST packets delayed in FDL corresponding to SM MTU

Packet switch

Cross coupling matrix1

s

s X PBS

1

s

1

s

1

s

s X PM

1

s

1

s

Packet/Circuitinput

Packet/CircuitoutputControl unit

GSTpackets

HCT/NCTpackets

FDL

FDL

SM

Inputs Outputs

21 Performance of SM class(Bjornstad et.al. ICTON 2004, JSAC 2006)

• GST packets introduce contention for SM packets, but…• High GST share implies fewer SM packets to be switched and buffered

– Lower SM PLR– Higher SM Delay

Performance varying GST traffic share

0,0E+00

1,0E-03

2,0E-03

3,0E-03

4,0E-03

5,0E-03

6,0E-03

0 20 40 60 80 100

GST traffic share (%)

PLR

0

0,5

1

1,5

2

2,5

Del

ay

PLRDelay

22 Performance and cost savings(Bjornstad et.al. ICTON 2004, JSAC 2006)

• GST performance (circuit)– No packet loss– Fixed delay (transmission delay + # nodes X FDL

• GST traffic not processed in intermediate packet switches– Transit traffic may follow the wavelength path if vacant capacity available– Transit traffic through the node does not need processing

• Less powerful packet switch required – Potential cost savings– Quantified in the ORION project

23

Experimental resultsFrom the OpMiGua testbed

24 OpMiGua testbedIngress

CorePolarisationcontrol

Fibre

25

Ingress

CorePolarisationcontrol

Fibre

OpMiGua testbed

Delay line

26

Experiments

• Polarization control experiments• Measuring packet loss of the SM class• Transferring video in GST or SM class

27

Polarization labeling (Tuft et.al. ICTON 2005 )

• Different from polarization multiplexing– Not simultaneous signals in SOPs

• Less sensitive to pol. Misalignment– Avoids inter-channel interference– 4.5 dB less power penalty @16 degrees

misalignment

• Experiments– Polarization misalignment sensitivity– Polarization control

SOP axesChannel 1 packets

Channel 2 packets

SOP axesChannel 1 packets

Channel 2 packets

PolMUX PolTDM

PolTDM

PolMUX

28

Measuring packet loss in OpMiGua

• Three node experimental network• GST class confirmed to have no packet loss

DelayGST λ1

GST detect

SM insert

Delay

SM insert

Delay

GST λ2

GST λ2

GST λ1

Ingress node

Corenode

Egressnode

SM dropSM drop /GST detect

gaps

29

SM PLR (ECOC 2006)

1.0E-05

1.0E-04

1.0E-03

1.0E-02

1.0E-01

1.0E+00

0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80SM load

Pack

et L

oss

Rat

io (P

LR)

GST load 0.29

GST load 0.39

GST load 0.49

GST load 0.58

SM PLR vs SM loads

SM Buffer overflowcauses steep risein SM PLR. Due to the CBR type of traffic, one or more packets are lost in each SM burst

For SM PLRs below 4x10-5, the total load (GST load + SM load) should be below 0.8. Total load 0.96 achieved in later exp. M. Nord et. al. PTL 2006

(Minimum SM PLR on the order of 10-6 due to error floor in PC generators. No error floor in PTL 2006 when using instrumental generators.)

30 Experimental setup video• Packet sources

– Uncompressed 270 Mb/s video-streaming over RTP/UDP/IP/Ethernet

– Packet generator and monitor

E/O GST interface

Logiccircuit

E/O SM interface

PBC

100 km SSMF

SOPSM

1550.5nm

PolTDM scheme SOPGST

SOPSM

Ingress

PBSFDL

(3.5 km) EDFA

SOPSM

PCSOPGST

power tap

Packet detector

PBC

SOPGST

1550.5 nm SOPSM

PC

PCSOPSM

O/E O/ECore Egress

25 km SSMF

Gigabit Ethernet packet generator

SOPGST

EDFA

PBS

SM interface

VideoEncoder

Logic/

Video monitorand recorder

packet counter

VideoDecoder

31

Two experiments on Gigabit Ethernet link• Experiment 1: Verify video transmission in GST class

without packet loss– Video quality should not be affected by SM packets

• Experiment 2: Video transmission in SM class.– Find how GST load degrades uncompressed video signals.

32

Experimental results, video in GST• GST always error free, independent of SM load

– Total load < 0.5 => SM PLR < 1 x 10-4 – Total load = 0.5 = saturation => SM PLR increases steeply to 2.3

%, 29 % @ load = 0.6

33Experimental results, video in SM• High error rate for total load >=0.5• Low error rate on SM for low GST loads• Error rate 5x10-4 gives visible errors in video!

0.5/5X10-4 0.7/1X10-3

0.8/4X10-1 0.9/8X10-1 1/~1

34

Further work

• Control and management of the OpMiGua hybrid network

• GMPLS compatible node and network design• Demonstration closer to a commercial product

35

Summary and conclusion• Migration from circuit to packet is a great challenge

– Applications like TV-broadcast and mobile puts very high demands on QoS– At its best: Low fixed delay for synchronization and a PLR < 10-8

– High cost efficiency

• OpMiGua hybrid circuit/packet network – GST packet transmission without packet loss, optionally with

synchronization– Processing of node transit traffic not required => Cost efficiency

• Experiments– Error free streaming of broadcast video in network with high total load– Load > 0.9 demonstrated with moderate SM PLR– Polarization labeling successfully demonstrated

• Further work– Management and control