Section 12 Traffic Management

1
Section 12 Traffic Management

• In this section
• Traffic measurements
• Quality of service
• Traffic management
• Congestion
• TCP congestion control
• Rate-based congestion control
• Integrated and Differentiated Services

2
Measuring Data Traffic

• Data rate amount of data sent per time interval
• Average data rate data rate over a significant
  time interval
• Burst a short sub-interval during which an
  unusually high data rate is present
• Maximum burst size maximum length of a burst
• Peak data rate maximum data rate during a burst

3
Measuring Data Traffic
4
Quality of Service (QoS) Parameters

• Reliability rate at which packet errors/loss
  occurs
• Throughput rate at which data passes through a
  fixed point.
• Delay Average amount of time taken to send
  packets from source to destination.
• Delay variation (jitter) Measure of difference
  between maximum and minimum delays for packets in
  same flow.
• Some applications are more sensitive to delay
  variation than delay
• Examples audio / video streams.

5
Traffic Management

• Fairness
• Quality of service
• May want different treatment for different
  connections
• What is more critical delay or loss?
• Reservations
• Traffic contract between user and network,
  specified at connection time.

6
Service Categories

• Real time
• Constant bit rate (CBR)
• Real time variable bit rate (rt-VBR)
• Non-real time
• Non-real time variable bit rate (nrt-VBR)
• Available bit rate (ABR)
• Unspecified bit rate (UBR)

7
CBR Constant Bit Rate

• Fixed data rate continuously available
• Tight upper bound on delay and delay variation
• Uncompressed audio and video
• Video conferencing
• Interactive audio
• Audio / video distribution and retrieval

8
rt-VBR Real-time Variable Bit Rate

• Time sensitive application
• Tightly constrained delay and delay variation
• rt-VBR applications transmit at a rate that
  varies with time
• Example compressed video
• At start, and periodically, send a complete image
  (a key frame)
• Otherwise, send differences from previous image
• Produces varying sized image frames

9
nrt-VBR Non-real-time Variable Bit Rate

• May be able to characterize expected traffic flow
• End system specifies
• Peak data rate
• Sustainable or average rate
• Measure of how bursty traffic is
• e.g. Airline reservations, banking transactions

10
ABR Available Bit Rate

• Application specifies peak data rate (PDR) and
  minimum data rate (MDR)
• Resources allocated to give at least MDR
• Spare capacity shared among all ABR sources

11
UBR Unspecified Bit Rate

• May be additional capacity over and above that
  used by CBR, VBR, and ABR traffic
• No resources dedicated
• For application that can tolerate some packet
  loss or variable delays
• e.g. TCP based traffic
• Packets forwarded on first-in, first-out basis
• Best efforts service

12
Congestion

• Congestion occurs when the number of packets
  being transmitted through the network approaches
  (or exceeds) the packet handling capacity of the
  network
• Congestion control aims to keep number of packets
  below level at which performance falls off
  dramatically
• Data network is a network of queues
• Generally 80 utilization is critical
• Finite queues mean data may be lost

13
Queues at a Node
out
in
14
Router Packet Handling

• Packets arriving are stored at input buffers
• Routing decision made
• Packet moves to output buffer
• Packets queued for output transmitted as fast as
  possible
• If packets arrive too fast to be routed, or to be
  output, buffers will fill and overflow.
• Can discard packets
• Can use flow control
• May propagate congestion through network

15
Interaction of Queues
16
Congestion Principles

• Usually occurs at a point of transition to
  reduced throughput.
• Occurs when the higher capacity part of a system
  is currently carrying more traffic than the lower
  capacity part can handle.
• Difference from flow control
• Flow control is one sender agreeing not to
  overflow one receiver at the endpoints of a
  transmission
• Congestion is usually caused by multiple senders,
  and occurs at an intermediate point in the
  network
• This makes congestion more difficult to detect,
  and to alleviate.

17
Idealized Performance

• Network can accept load up to its capacity,
  regardless of throughput.
• If load is increasing faster than throughput,
  packets will have to queue up
• This increases delay
• Additional load will be delivered at capacity
  throughput rates.
• Load accepted at beyond capacity throughput rates
  will increase the queue size.

18
Idealized Performance Throughput
19
Idealized Performance Delay
20
Practical Performance
21
Practical Performance

• Ideal assumes infinite buffers and no overhead
• Buffers are finite
• Overhead occurs in exchanging congestion control
  messages

22
The Congestion Control Paradox

• When congestion occurs, the problem is that there
  are too many packets in the network
• If packets are trashed, senders will likely
  resend them, along with new packets.
• Result increased congestion
• If one node sends out messages to announce it is
  congested, then it increases the number of extra
  overhead packets in the network.
• Result increased congestion
• If one node asks its neighbours to slow down,
  then the output queues of the neighbouring nodes
  will start filling up.
• Result increased congestion

23
Congestion Control

• Implicit
• No action taken
• It is assumed senders will notice evidence of
  congestion and deal with it themselves.
• What can senders do?
• Slow rate of packet sending
• Increase timeout length for sent packets
• Explicit
• Various mechanisms to announce or alleviate
  congestion, taken by intermediate network notes.

24
Implicit Congestion Detection

• What are the signs of congestion?
• Increased transmission time
• Packets spend more time in queues that are
  longer delay increases
• Disappearance of packets
• On a fibre-based network (or ones with data link
  error control), disappearance of packets can be
  interpreted as a sign of congestion.
• Sending timers expire more frequently.

25
Explicit Congestion Signaling

• Network alerts end systems of increasing
  congestion
• End systems take steps to reduce offered load
• Backwards
• Congestion avoidance in opposite direction to
  packet direction
• Forwards
• Congestion avoidance in same direction as packet
  direction
• Receiver can use flow control to slow sender, or
  otherwise notify sender of congestion.

26
Backpressure

• If node becomes congested it can slow down or
  halt flow of packets from other nodes
• May mean that other nodes have to apply control
  on incoming packet rates
• Propagates back to source
• Can restrict to logical connections generating
  most traffic
• Used in connection oriented protocols that allow
  hop by hop congestion control (e.g. X.25)

27
Choke Packet

• Control packet
• Generated at congested node
• Sent to source node
• e.g. ICMP source quench
• From router or destination
• Source cuts back until no more source quench
  message

28
Categories of Explicit Signaling

• Binary
• A bit set in a packet indicates congestion
• Credit based
• Indicates how many packets source may send
• Common for end to end flow control
• Rate based
• Supply explicit data rate limit
• e.g. ATM

29
TCP Congestion Control

• Detecting congestion the main symptom is
  transmission time-outs.
• TCP assumption Data links are now generally
  reliable, and therefore time-outs indicate
  congestion.
• TCP maintains a congestion window that is
  separate from the receivers flow control window.
• Sender can only send as many bytes as the smaller
  of the two windows.

30
Congestion Policy

• Set initial size of congestion control window at
  maximum single segment size agreed on at
  connection time.
• Set initial threshold at 64K bytes.
• For each segment acknowledged before timing out
• If below threshold, double size of congestion
  control window.
• If above threshold, add one more maximum segment
  size to congestion window.
• When a segment times out
• Cut threshold in half from current window size.
• Reset window size to initial value.
• Result
• Exponential growth up to threshold.
• Linear growth afterward.
• Called TCP slow start

31
TCP Slow Start
32
Rate-based Congestion Control

• Regulate rate at which sender can inject packets
  into network
• A packet must match up with (and remove) a token
  before entering network.
• Tokens added to bucket at rate r.
• At most b tokens can accumulate in bucket tokens
  overflow and are lost after that
• Bucket size b controls burstiness
• Max. number of packets entering network in t, t
  d is b ? dr

tokens arrive at fixed rate
storage for up to b tokens
bucket of tokens
packet waiting area
to network
33
Traffic Management and Congestion Control
Techniques

• Queue management
• Connection admission control
• Selective packet discard
• Traffic shaping
• Use the token bucket scheme for rate-based
  congestion control.

34
Queue Management

• First in, first out (FIFO)
• Packets are processed in the order of their
  arrival.
• When the queue is full, packets are discarded
• Priority queues
• Packets are classified and directed to one of a
  number of queues with various priorities
• Packets are processed from highest priority queue
• Packets in lower priority queues not processed
  until all higher priority queues are empty
• Weighted fair queues
• Packets are classified as with priority queues.
• Process up to n packets from priority n queue
• Go to next queue if empty.
• Guarantees all queues are served and prevents
  starvation.

35
Connection Admission Control

• First line of defense
• User specifies traffic characteristics for new
  connection by selecting quality of service
  parameters.
• Network accepts connection only if it can meet
  the demand
• Traffic contract
• Peak data rate
• Packet delay variation
• Sustainable data rate
• Burst tolerance
• Monitor above parameters, and if traffic does not
  conform to the contract, packets are liable to be
  discarded.

36
Integrated Services Architecture

• Acronym ISA
• Standards currently under development by IETF
• Base document in RFC 1633
• Categories of traffic
• Inelastic constraints on throughput, delay,
  jitter, and packet loss
• Elastic can adjust to changes in network
  conditions
• Varying tolerances for changes in above factors
• E-mail sensitive to loss, but not delay
• FTP file transfer sensitive to throughput, but
  not jitter

37
ISA Services

• Guaranteed service
• Assured data rate
• Upper bound on queuing delay
• No queuing losses
• Controlled load
• Similar to guaranteed service, except that
  constraints are only expected to be met for a
  high percentage of packets instead of all
  packets.
• Best effort
• No quality of service parameters applied to
  traffic.

38
Elements of ISA

• Routing algorithm
• As an alternative to delay, quality of service
  can be used to weight graph edges for OSPF
• Admission control
• For any service other than best effort, a
  reservation must be made using the RSVP protocol
  (RFC 2205)
• Queuing Discipline
• Multiple output queues with fair selection for
  transmission
• Each flow of inelastic traffic can be queued
  separately
• Discard Policy
• Policy for which packets to discard when a queue
  is full.

39
ISA Router Architecture
Reservation Protocol
Admission Control
Routing Protocols
Routing Database
Traffic Database
Management Agent
QoS queues
Classification and Route Selection
Packet Scheduler
Best effort queue
40
ISA Issues

• The RSVP protocol requires each router along a
  path to store quality of service (QoS)
  information about each data flow.
• Scalability puts additional burden on routers
• Types of service are limited
• Architecture fixes the types of services that can
  be provided guaranteed, controlled, or
  best-effort.

41
Differentiated Services (1)

• Abbreviations DiffServ, DS
• Objective move the classification of services
  to the edge of the network instead of the
  routers.
• Uses the current IP TOS (type of service) field
• Data source, or first router, is responsible for
  filling in this value to match various service
  classes that can be offered.
• Intermediate routers can then use this field for
  routing decisions, instead of having to store
  information about each flow.

42
Differentiated Services (2)

• First 6 bits of TOS field used to identify
  service classes on a per hop basis
• Currently, 3 classes have been defined that are
  similar to those in ISA
• Default effort (DE) no special treatment
• Expedited forwarding (EF) effectively results
  in a virtual circuit
• Assured forwarding (AF) similar to ISAs
  controlled load, where packets will meet service
  guarantees unless the traffic profile is exceeded
  at a node.

43
Differentiated Services Router
Meter
Marker
Shaper/ dropper
Classifier

• Meter monitors traffic for profile conformance
• Marker may downgrade service class for traffic
  that violates profile
• Shaper and/or dropper reforms traffic and/or
  drops packets