A Summary of Engineering Rules for ATM Network Dimensioning and QoS

1
A Summary of Engineering Rules for ATM Network
Dimensioning and QoS

• Dr John A. SchormansQMWLondonUK

2
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.

3
NP parameters
CTD is the total delay experienced by a cell
traversing a network CDV is a measure of the
variation of a cells delay in crossing the
network CLR proportion of lost cells These are
all quantifiable, in the long run, at least
theoretically.
4
EXPERT project methodology

• Experiment with real networks and applications,
  where available,
• use simulation otherwise.

5
EXPLOIT Configuration Basel
Whitetree
ATMLightRing
pc- atm01
pc- atm02
PC MM2
TA
TA
002
2/2
TA
11
Ascom ISS
Olicom OC-9100
Cisco LS1010
003
1
001
2/1
1-10
Ring 1
PC MM3
2/3
000
Ring 2
TA
12
2/4
2
3
a
clock

pc- atm03
TA
TA
ASX-200 D1
T
187
B12
Italtel / Lucent APON
TA
Siemens EWSXpress 36190
T
pc- atm04
Wandel Goltermann
B14
TA
T
B22
x2
ATM 100
381
ATM100
T
TA
Splitter
B24
V
V
V
AAU 5
V
115
175
GN Nettest
B4
B3
B1
B2
IW95000
x2
IW95000
WS jupiter
Alcatel
TA
ASX-200 D4
clock
8
16
17,19,35
254
32
C1 C2 C3 C4
A1
A8643
B1 B2 B3 B4
HQA
9
TA
S
24
Lucent RUM
D3
Philips LaTEX
WS saturn
Alcatel
FORE ASX-200
TA
18
4
28
A2
A8640
x2
HQA
TA
R
22 23
30
11
RUM,9
D4
10
20
34
26
31
WS jungfrau
D2 A3 A4
D1
TA
ATM serial 155 Mbit/s, optical
cellbased parallel 155 Mbit/s, electrical
cellbased STM1 155 Mbit/s, electrical STM1
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, electrical 25 Mbit/s , Twisted
pair policed inputs non ATM 2 Mbit/s
CE Ethernet Frame Relay, 2 Mbit/s
WS saentis
TA
PC
TA

TA
PC
EWSXpress, 175
11
1
5
8
Ascom AAU
4
10
7
H261
9
12, 2
N-ISDN
TA
Videostudio
g\aspa\platform\etb_conf\expert\vers4_6.ppt/18.8.
97
6
(No Transcript)
7
Different forms of buffer behaviour ...
8
(No Transcript)
9
(No Transcript)
10
(No Transcript)
11
X
X the buffer length in cells C the rate at
which the buffer is actually being served
12
overflow probability vs buffer length for ON-OFF
source
200
400
600
800
1000
1
0.1
overflow probability
0.01
0.001
0.0001
buffer length, X
13
time priorities, buffer management 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 buffer are served before
cells from priority 2.
14
time priorities, implications for CAC
- For rt-traffic CAC needs to check that a) the
new connection is acceptable to the
rt-traffic, b) that its 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).
15
time priorities, linearity of EB approaches
- Effective b/ws 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 EBs high enough.
16
(No Transcript)
17
Shaping
cell departing at lt the line rate
cells arriving at the line rate, say 155Mbit/sec
To shape 155Mbit/sec down to 25Mbit/sec
back-to-back cells depart 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.
18
This is shaping applied to ftp traffic
19
This is shaping applied to ftp traffic. It is the
time taken to transmit a 10Mbyte binary data
block
20
(No Transcript)
21
multi-media terminal, ISABEL
This application was intended for the
interconnection of audiences, and has was
developed for the RACE ACTS summer schools. -
Tele-education / training - Telework -
Telemeeting Based on SUN workstations, this
means that the IP protocol stack uses all the
available bandwidth during an ON period.
22
(No Transcript)
23
ATM output capacity was 149Mbits/sec, reducing
the shaper rate down from 100Mbits/sec to
4Mbits/sec increases the number of acceptable
simultaneous connections
60
24
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).

25
Minimum QoS to CLR mapping
application cell loss ratio
26
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.

27
Early packet discard (EPD)
this whole packet is discarded, as it arrived to
a queue length gt than the threshold
buffer size
Discard
threshold
buffer
occupancy
t
t
t
t
t
1
2
3
4
5
time (cell slots)
Individual cell from packet
th
t
Arrival time of i
packet
i
28
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.
29
TCP over ATM
end-to-end reactive flow control
TCP (tx)
TCP (rx)
potential problem with nested flow control loop
s
ABR features
ATM cell-based flow control
Else UBR has cell-based transmission
with no flow control
30
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 TCPs slow-start
  mechanism could be redesigned to take advantage
  of guaranteed ATM b/w.

31
For TCP over ATM, the results of certain
studies showed that
32
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 TCPs slow start algorithm
  is unnecessary.

33
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.

34
Recommendations, CLR

• In the past CLR of 110-9 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.