+ All Categories
Home > Documents > A Summary of Engineering Rules for ATM Network Dimensioning and QoS

A Summary of Engineering Rules for ATM Network Dimensioning and QoS

Date post: 03-Feb-2016
Category:
Upload: moral
View: 31 times
Download: 0 times
Share this document with a friend
Description:
Dr John A. Schormans QMW London UK. A Summary of Engineering Rules for ATM Network Dimensioning and QoS. Quality of Service (QoS) is a user-related concept, and refers to essentially subjective quantities: - picture flicker, - annoying clicks in sound etc, - time to transfer data. - PowerPoint PPT Presentation
34
QUTE’98 Workshop A Summary of Engineering Rules for ATM Network Dimensioning and QoS Dr John A. Schormans QMW London UK
Transcript
Page 1: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

A Summary of Engineering Rules for ATM Network Dimensioning and

QoS

Dr John A. SchormansQMW

LondonUK

Page 2: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

QoS and NP

• Quality of Service (QoS) is a user-related concept, and refers to essentially subjective quantities: - picture flicker, - annoying clicks in sound etc, - time to transfer data. QoS is the overall impression the user obtains from the system, from the physical layer to the application layer.

• The network operator should seek to guarantee the measurable network performance (NP) parameters: CTD, CDV and CLR.

Page 3: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

NP parameters

CTD is the total delay experienced by a cell traversinga network

CDV is a measure of the variation of a cells delayin crossing the network

CLR proportion of lost cells

These are all quantifiable, in the long run, at least theoretically.

Page 4: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

EXPERT project methodology

• Experiment with real networks and applications, where available,

• use simulation otherwise.

Page 5: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

SC

SC

34 M

SC

H261 12, 2

7

8 111

AscomAAU

78

116 10

5

4LucentRUM

202223

24

28

16 17,19,35

PhilipsLaTEX

PC

PC

TA

TA

9254

10PC TA

18

g:\aspa\platform\etb_conf\expert\vers4_6.ppt/18.8.97

ASX-200D4 32

AAU 5

EWSXpress,1755

V

Italtel /LucentAPON

C1 C2 C3 C4

A2FOREASX-200

SC

D2 A3 A4

SiemensEWSXpress 36190

ASX-200D1

APON,Tb

RUM,9 D4

115175

D1

ATM:

serial 155 Mbit/s, optical cellbasedparallel 155 Mbit/s, electrical cellbasedSTM1 155 Mbit/s, electricalSTM1 155 Mbit/s, optical (monomode)STM1 155 Mbit/s, optical (multimode)Taxi 100 Mbit/s, optical (multimode)serial 1,25 Gb/s, optical (cellbased) PDH 34 Mbit/s, electrical25 Mbit/s , Twisted pairpoliced inputs

non ATM:2 Mbit/s CEEthernetFrame Relay, 2 Mbit/s

B1B2B3B4

9

EXPLOIT Configuration Basel

381ATM100

IW95000

x2

pc-atm01TA

OlicomOC-9100

CiscoLS1010

AscomISS003

002

001

000

pc-atm02

TA

1-10

2/22/1

2/4

D3

A1

SwissWAN

pc-atm03

WSsaentis

pc-atm04

WSjungfrau

WSjupiter

WSsaturn

TAPCMM3

TAPCMM2

187

3426 31

2/3

HQA

HQA

TA

TA

S

R

A8643

A8640 x2

x2

N-ISDN TA

SCTATVS

TVR Splitter

TA

TA

TVR

TVR

IW95000

ATM 100

TA

TA

TA

TA

TA

TA

4

30

11

12

1

2 3

Ring 1Ring 2

ATMLightRing Whitetree

TB14T

TV V V

TB12

B24

B22

B4 B3 B1B2

a

clock

clock

Videostudio

Wandel & Goltermann

GN Nettest

Alcatel

Alcatel

Page 6: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

BaselLeidschendam

738 km

TestEquipmentAmsterdam ZurichCologne

BaselMontrealOttawa

Tele-Teaching

Tele-

Teaching

4575 km

TestEquipment

Pennant

PointHamburg ZurichCologne

Page 7: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Different forms of buffer behaviour ...

1

N

o/p capacity ‘C’

‘C’

‘C’

time

time

no. of arrivals per unit time

no. of arrivals per unit time

cell scale queueing, short buffers

burst scale queueing, long buffers

Page 8: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

queuesize

(a) all streams out of phase

queuesize

(b) two streams in phase

queuesize

(c) all streams in phase

Page 9: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

CLP [ e - 1 - 2/3 ]x

CLP (exp[-2(1-).(1 + (1-))/3])x

1E-10

1E-09

1E-08

1E-07

1E-06

1E-05

0.0001

0.001

0.01

0.1

1

0 10 20 30 40

Buffer capacityC

LP

Poisson 95%

Poisson 75%

Poisson 55%

X = the buffer length in cells

Page 10: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

R

Cell rateeffective bandwidth

EB

Time

Time

Cell rate

ON OFF

Page 11: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

1 - 1/(Eon(R-C))CLP __________________

1 - 1/(Eoff.C)

X

X = the buffer length in cellsC = the rate at which the buffer is actually being served

Page 12: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

overflow probability vs buffer length for ON-OFF source

0.0001

0.001

0.01

0.1

1200 400 600 800 1000

buffer length, X

ove

rflo

w p

rob

abil

ity

Page 13: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

time priorities, time priorities, buffer management buffer management strategy for integration of rt-VBR and nrt-VBR strategy for integration of rt-VBR and nrt-VBR

priority 1 cell buffer

realtime traffic

priority 2 cell buffer

non-realtime traffic

.

Cells from pr 1 bufferare served before cellsfrom priority 2.

Page 14: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

time priorities, time priorities, implications for CAC implications for CAC

priority 1 cell bufferrealtime traffic

priority 2 cell buffernon-realtime traffic

.

- For rt-traffic CAC needs to check that a) the new connection is acceptable to the rt-traffic, b) that it’s effect on the nrt-traffic will not violate the guaranteed NP of the nrt-traffic.

- For nrt-traffic CAC needs part (b) only (due to priority mechanism).

Page 15: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

time priorities, time priorities, linearity of EB approachesof EB approaches

priority 1 cell bufferrealtime traffic

priority 2 cell buffernon-realtime traffic

.

- Effective b/w’s have been proposed for CAC, and EXPERT experiments have been oriented towards testing the linearity of such a scheme.

- Importantly: linearity was found in experiments, but also significant deviation from linearity. Such deviation can be compensated for by making the EB’s high enough.

Page 16: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

0.01

0.1

1

10

100

1000

46000 47000 48000 49000 50000 51000 52000 53000Cell Slots

Inst

anta

neou

s B

andw

idth

Mbi

ts/s

ec

IP data traffic with peak rate = 155.52Mbits/sec and mean rate = 22.5 Mbits/sec

Page 17: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Shaping

cells arrivingat the line rate, say 155Mbit/sec

cell departing at < the line rate

To shape 155Mbit/sec down to 25Mbit/sec ‘back-to-back’ cellsdepart separated by 6 cell slots (80% of them), or 7 cell slots (20%of them), theoretically! In practice the interface between ATM and the underlying SDH caused smaller gaps in practice.

Page 18: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

0

20

40

60

80

100

120

140

160

0

500

1000

1500

2000

2500

3000

3500

Time in Cell Slots

Inst

anta

neou

s Ban

dwid

th

No Shaping

25 Mbits/s

10 Mbits/s

This is shaping applied to ftp traffic

Page 19: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

1

10

100

1000

0.1 1 100 1000Shaper Peak Rate in Mbits/s

Tim

e in

Sec

onds

Pea

k B

andw

idth

155.

52 M

bits

/s

This is shaping applied to ftp traffic. It is the time taken to transmit a 10Mbyte binary data block

Page 20: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Real Traffic Markov Models

Real Traffic Markov Models

Page 21: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

multi-media terminal, ISABEL

This application was intended for the interconnection ofaudiences, and has was developed for the RACE & ACTSsummer schools.

- Tele-education / training

- Telework

- Telemeeting

Based on SUN workstations, this means that the IP protocolstack uses all the available bandwidth during an ON period.

Page 22: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

0

200

400

600

800

1000

1200

1400

1600

1800

2000

0 10 20 30 40 50 60 70 80 90 100Shaping Rate in Mbits/s

Shap

er Q

ueue

Siz

e

This is for real-time multi-media traffic, so thequeueing delays must be minimal.

Page 23: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

60

ATM output capacity was 149Mbits/sec, reducing the shaper rate down from 100Mbits/sec to 4Mbits/sec

increases the number of acceptable simultaneous connections

0

10

20

30

40

50

0 10 20 30 40 50 60 70 80 90 100

Shaper Rate in Mbits/s

Nu

mbe

r of

Sou

rces

Queue = 100Queue = 50NDi/D/1M/D/1

This is for real-time multi-media traffic, so thequeueing delays must be minimal.

Page 24: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

subjective assessment

• In experiments Audio transmission supported by ATM suffered from imperfections (clicks, loss of sound), as a result of cell loss.

• A higher cell loss probability was more acceptable to users of narrowband telephony than high quality audio; this is due to the expectation of higher quality from the latter.

• Testers of HQ audio expected CD quality sound, but users of a narrowband telephony circuit found their conversation still understandable when 8% of cells were lost (if the distribution of cell losses was uniform).

Page 25: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Minimum QoS to CLR mapping

H621 videoconferencing

1.85 x 10-5

384 kbit/sec HQ audio 4.57 x 10-5

64 kbit/sec audio 3.42 x 10-4

Isabel m/mediaterminal

1 x 10-4

application cell loss ratio

Page 26: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Shaping - conclusions

• For the user, adding a shaper: - increases the complexity of the terminal equipment, - adds delay to the end-to-end connection (particularly a problem for AAL5 connections, as the full PDU is needed at the receiver. - reduces the peak rate of the connection, which should reduce the cost,

• For the network operator:- shaping reduces the ‘effective bandwidth’ of the sources, so increasing the number that can be safely multiplexed.

Page 27: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Early packet discard (EPD)

time (cell slots)

buffer size

Discardthreshold

Individual cell from packet

bufferoccupancy

t1 t2 t3 t4 t5

ti Arrival time of ith packet

this whole packetis discarded, as itarrived to a queue length > than the threshold

Page 28: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Assumptions:

1. cells from a packet arrive instantaneously,2. the arriving packets have a homogeneous packet size,

N cells,3. output link rate is constant, thus cell service time is

deterministic, 1 cell per time unit,4. packet arrivals are i.i.d.

Page 29: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

TCP over ATM

TCP (tx) TCP (rx)

end-to-end reactive

flow control

ATM cell-based flow control

ABR features

Else UBR has cell-based transmission with no flow control

potentialproblem with nested

flowcontrolloops

Page 30: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Engineering rules for ABR

• The burstiness of the rt-traffic has a significant effect on the ABR throughput: the more bursty the worse the throughput.

• For ATM over mobile or over satellite the TCP rate reduction mechanism has been found to be less effective than the Explicit Rate ABR feedback mechanism. The ABR mechanism offers greater precision and may be able to prevent congestion rather than simply reacting to it.

• Carrying TCP over CBR connections was found to be an inefficient method of transporting TCP packets.

• It was suggested that TCP’s “slow-start” mechanism could be redesigned to take advantage of guaranteed ATM b/w.

Page 31: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

UBR with EPD same performance asstandard packet networks

UBR without EPD worse performance thanstandard packet networks

ABR worse performance than UBR without EPD

For TCP over ATM, the results of certain studies showed that:

Page 32: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Recommendations

• It is difficult for TCP flow-control algorithms to work well in an environment where delays are large, e.g. in satellite or mobile ATM, as congestion notification is often outdated by the time it is received. Flow-control here is better provided by ABR, as this can respond faster.

• Carrying TCP over CBR connections does not contribute to efficient network operation.

• ATM can often provide a near guarantee of frame rate transmission, so TCP’s slow start algorithm is unnecessary.

Page 33: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Recommendations, CTD, CDV

• CTD is difficult to specify in a traffic contract, and, when building a network, the delay should be made as small as possible, as the cost to the network operator of transporting the cells across the network increases with the latency in the system: reducing the latency in the system promotes a greater turnover of customers and hence increased revenue.

• rt-VBR circuits are difficult to manage when statistically multiplexed. Any “burst scale” component fills the small buffers causing low quality of service. Very low utilisation factors (less that 4%) have been found essential for unshaped traffic. Therefore shaping and/or peak rate allocating all real-time circuits may be recommended to guarantee end-to-end quality of service for CDV sensitive services.

Page 34: A Summary of Engineering Rules for ATM Network Dimensioning and QoS

QUTE’98 Workshop

Recommendations, CLR

• In the past CLR of 1*10-9 have been recommended for all traffic over ATM networks. This is now very debatable, as the application and its purpose is most important in assessing the required CLR.

• The CLR of N-ISDN speech is quite high, 1 in 3000 cells is acceptable and with even 15% cell loss the conversation is still understandable.

• For High Quality CD sound the user expects to have the purest reproduction possible, else the user would much prefer to use a “music system”. So the CLR needs to be much lower: experiments used an application able to correct 1 cell in a block of 8, therefore the maximum CLR was found to be 1 in 20,000. Assuming no FEC was present, then the CLR requirement would be less than 1 in 17,000,000.

• Video was acceptable with higher CLR ratio e.g. 1/10,000.


Recommended