Traffic and Congestion Control in ATM Networks

1
Chapter 12

• Traffic and Congestion Control in ATM Networks

2
Overview

• Open Loop flow control?
• No feedback, traffic shaping
• Closed Loop control?
• Feedback used to do source to destination
  control.
• Absolute peak control?
• Allocating peak bandwidth for each channel. No
  statistical multiplexing possible

3
Overview

• Why is congestion control difficult in ATM
  networks?
• High bandwidth, short cells, more cells in
  transit
• Multiple QoS types on same network

4
Model
tf
Source
Dest
5
Congestion Control Limits

• Round-trip propagation delay
• Forward path and feedback message delay
• delay-bandwidth product
• The number of cells which will be transmitted
  before congestion control can be effective

6
Examples

• 45Mbps, 9.4usec per cell, 106,000 cells/sec,
  1000Km, 5usec/km
• delay 5msec mt530 cells
• 2.4Gbps, mt28,000 cells

7
Reassembly of CBR Cells
8
Traffic Descriptors

• Source Traffic Descriptor
• Peak Cell Rate (PCR)
• Sustainable Cell Rate (SCR)
• Maximum Burst size
• Minimum Cell Rate
• Connection Traffic Descriptor
• Cell Delay Variation Tolerance

9
QoS parameters

• Cell Delay Variation CDV
• Maximum Delay CTD
• Cell Loss CLR

10
Delay Probability Density
11
VCCs and VPCs
1,2 QoS differs from 3,4,5
Group according to QoS, set CIR for VPC to sum of
VCCs or to something less (statistical
Multiplexing)
12
Location of UPC
13
Generic Cell Rate Algorithm
I10, TAT10, ta(1)8 CDVTL3 TAT20, ta(2)28
TAT 38, ta(3)30, nonconforming
14
Generic Cell Rate Algorithm
I10, LCT8, ta(1)8, X0, L3 X10, LCT8,
ta(2)28, (10-(28-10)) lt 0 X 10, LCT28,
ta(3)30, (10-(30-28))8 8 gt 3 nonconforming
15
GCRA
16
Cell Arrival
17
Leaky Bucket Algorithm for traffic shaping
18
Actions of UPC Function
19
Violation Tagging

• Violation tagging
• Cells are divided between high priority and best
  effort, over contract cells.
• Example- voice, divide 4 MSB into non-tagged
  stream, 4 LSB tagged stream
• High priority video frames sent in own stream.

20
Traffic Shaping
21
ABR Service
22
EFCI (Explicit Forward Congestion Indication)

• Resource Management Cells are sent periodically
  (every 32 cells default)
• The EFCI bit is set if More than one buffer is
  occupied.
• The switch can also generate a RM cell in the
  backward direction or mark BRM cells as they go
  by in the opposite direction of the congested
  flow.
• The switch can also modify the ER (Explicit Rate)
  field of Backward or Forward RM cells

23
Power Function
delay
Max Power
Throughput
24
Maximizing Power

• Power throughput/delay
• for M/D/1

25
Queue Length
Packets passing through switch will be set 50
of the time
26
Rule

• Use rule that if 50 of bits are set, reduce rate

27
Oscillations

• Two methods proposed to dampen oscillations
• Base action on average queue level, not current
  level
• Reduce window size at a smaller rate

28
Maximum Rate
29
Rate Based ATM Forum

• EPRCA?
• Enhanced Proportional Control Algorithm
• Rate based chosen over credit based approaches
• Credit based requires Per VC queuing
• Link by link flow control

30
EPRCA

• VC transmits at Allowed Cell Rate (ACR)
• Rate is reduced by ADR after each cell
• Every Nrm cells, generate RM cell in backward
  direction
• When received from Destination, increment rate by
  NrmADR
• Rate is reduced if RM cell is lost

31
PT Field

• 000 User Data, no Congestion, SDU 0
• 001 End of AAL5 packet
• 010 User Data, Congestion, no end
• 011 User Date Congestion, End of AAL5 packet
• 100 Segment OAM flow
• 101 End to End OAM
• 110 RM Cell
• 111 Reserved

32
Variations in Allowed Cell Rate
RDFRate Decrease Factor1/4 RIFRate Increase
Factor1/16 CICongestion Indication NINo
Increase
33
Flow of RM cells in ABR connection
Source sets ERdesired rate NINo Increase,
CICongestion Indication
34
Credit Based Flow Control
35
Credit Based Flow control eliminates overflow
36
Per VC queue necessary for Credit based approaches
37
Hybrid Credit/Rate Based
38
Rate Based Schemes require larger buffers
39
Credit based approaches keep buffers full

• Since I allocate credits whenever buffers are
  available, they remain full generally.
• What does this do to delay?

40
Simulation Environment

• ATM simulators are used to test configurations,
  traffic patterns, flow control and many more
  parameters without putting the money and time
  into creating a real network.
• NISTs ATM simulator is used around the world to
  further research on ATM.

Brigham Young University
Networked Computing Lab
41
Rate-based Example
VBR Source Link Rate
ABR Source Link Rate
Backbone Link Rate
Brigham Young University
Networked Computing Lab
42
Credit-based Example
VBR Source Link Rate
ABR Source Link Rate
Backbone Link Rate
Brigham Young University
Networked Computing Lab
43
Simulations

• Used three standard network configurations to
  test known flow control problems.
• Added VBR background traffic using MPEG-2 traces
  to simulate a dynamic environment.
• Ran simulations in both MAN and WAN environments
  (100 km and 1000 km links respectively).
• Performance was evaluated in terms of throughput,
  buffer occupancy, burstiness, fairness, and fast
  convergence.

Brigham Young University
Networked Computing Lab
44
Configurations

• Fast Convergence Tests for the time required
  for ABR sources to converge to the fair
  allocation of available bandwidth.
• Parking Lot Tests if all sources receive a fair
  share of the bandwidth, even though some may be
  routed through more switches than others.
• Upstream Bottleneck Tests to see if available
  bandwidth will be reallocated when a source is
  unable to use it.

Brigham Young University
Networked Computing Lab
45
100
150
100
50
Fast Convergence
Brigham Young University
Networked Computing Lab
46
100
100
150
100
50
Parking Lot
Brigham Young University
Networked Computing Lab
47
Should get 33
Should get 66
50
Upstream Bottleneck
Brigham Young University
Networked Computing Lab