Traffic Engineering of WiMAX

1
Traffic Engineering of WiMAX

• Villy B. Iversen

2
Master Theses

• Traffic and Service Engineering
• (Yahya Sethi)
• WiMAX Network Planning
• (Qi Zhang)
• Quality-of-Service in WiMAX Access Network
• (Ahmed Imran Najam)

3
Presentation Overview

• What is WiMAX ???
• Introduction to the four service Classes
• Present an analytical model
• Performance evaluation using simulation
• Conclusion

4
Introduction to WiMAX

• WiMAX acronym !!!
• Based on IEEE802.16
• Aims at Conformance Interoperability
• Uses a PMP toplogy with a controlling BS and
  attached SS

5
Service Classes

• UGS
• Fixed size data grants sent periodically
• Suitable for constant bitrate e.g. Voice
• rtPS
• Unicast polls sent periodically
• For variable bitrate traffic e.g Video
• nrtPS
• Unicast polls not necessarily periodic e.g FTP
• BE
• Contention slots
• Designed for Internet Traffic

6
Analytical Model

• Network to provide QoS but does not specify the
  means of doing that
• Scheduling plays an important role by providing
  differential treatment
• Propose an analytical model according to their
  QoS requirements
• The performance metric Packet Delay
• Mean Packet delay Queuing Access delay

7
Analytical Model (cont ...)
8
Analytical Model (cont ...)

• Mixture of Priority queues and FIFO
• Four queues correspond to the four service
  classes
• Each queue being served using FIFO
• Pre-emptive non-pre-emptive priority scheme
• Model Corresponds to GPCC scheme
• Essence of the proposed model

9
UGS Queuing Delay

• Mean Waiting time contains two contributions
• Packets already undergoing service
• Packets already ahead of the arriving packet
• Mean Waiting time

10
rtPS Queuing Delay

• Mean Waiting time contains three contributions
• Packets undergoing service
• Packets ahead of the arriving packet
• Arrival of higher priority packets (UGS)

11
rtPS Queuing Delay (cont ...)

• Mean Waiting time

12
nrtPS Queuing Delay

• Mean Waiting time contains three contributions
• Packets already undergoing service
• Packets already waiting in the queues to get
  service higher priority arrival

13
nrtPS Queuing Delay (cont ...)

• Pre-emptive priority of higher priorty classes
• Mean Waiting time

14
BE Queuing Delay

• Mean Waiting time contains three contributions
• Packets already undergoing service
• Packets already waiting in the queues to get
  service higher priority arrival

15
BE Queuing Delay (cont ...)

• Pre-emptive priority for UGS and rtPS
• Mean Waiting time

16
Simulation

• Behaviour of service classes in varying network
  load
• Effect of Fragmentation/Concatenation
  Piggybacking
• Size of the contention window
• Queuing model verification

17
Scenario 1 Behaviour of service classes

• Each SS supporting one of the service class
• FTP used as application traffic
• Each FTP session corresponds to a load 0.16Mbps
• Network loaded by adding more SSs

18
Scenario 1 (cont ...)

• Download Response time for the four service
  classes when the network is loaded with 7.2Mbps
  of application traffic

19
Scenario 1 (cont ...)

• UGS and rtPS service classes show a load
  independant delay
• UGS rtPS have same polling interval but
  different timings
• Polling characteristics of rtPS makes it suitable
  for variable bit rate
• UGS and rtPS exhibit constant jitter
• BE service is network load load dependant
• High load results in scarcity of free contention
  slots
• Repeated collisions results in buffers getting
  full
• No guarantees in jitter

20
Scenario 1 (cont ...)

• nrtPS exhibits a faulty behaviour
• Uses only the contention slots
• Problem in the MAP SEND state of BS process model
• Owing to faulty behaviour it is excluded in
  further simulations

21
Scenario 2 Effect of Piggybacking

• Despite effective link capacity being 35Mbps BE
  response time saturated at 7.2 Mbps
• Saturation Previous request not being fullfiled
  until the time for next
• Piggybacking Sending the next request with
  allocated grant
• Piggybacking not applicable to UGS and rtPS class

22
Scenario 2 (cont ...)

• Provides better utilization of network
• Reduction in contention cycles decreases the
  response time
• UGS and rtPS remains unperturbed by it

23
Scenario 3 Effect of Fragmentation
Concatenation

• Concatenation applicable to all the four service
  classes
• Concatenation Request made for multiple packets
  allows reducing overhead
• Fragmentation allows better network utilization
• BS allocates a partial or full grant depending on
  network condition
• SS utlizes the partial grant and sends the
  fragment of data
• Without fragmentation grant would have gone wasted

24
Scenario 3 (cont ...)

• Medium loaded network of 15 SSs
• Load varied by
• changing file
• inter request

25
Scenario 3 (cont ...)

• Concatenation allows to have an arbitrary MAP
  duration
• Fragmentation Disabled Network gets saturated at
  30
• With increase in load BS sends partial grants
• Partial grants go wasted
• Queue size builds up
• Response time saturates
• Fragmentation Enabled
• Partial grants utilized by SS in sending
  fragments of data
• Queue size remains bounded

26
Scenario 4 Effect of Backoff Start Value

• Study of contention window size.
• Backoff start and end value define the contention
  window
• Backoff start value defines the size of initial
  deference window
• Larger Backoff start value seems plausible
  solution. BUT !!!!!!!!
• Number of SS is remained fixed and load is varied
  by inter-request time

27
Scenario 4 (cont ...)

• For low loads larger backoff value desirable
• Small download response time due to reduced
  collsions and retranmissions
• For heavy load all the three backoffs give
  similar response time
• Decrease delay due to reduce collisons offset by
  the queuing delay

28
Scenario 5 Evaluation of Queuing Delay

• Queueing delay obtained from CSF model and
  proposed model are compared
• Voice Traffic chosen as an Application traffic
• Voice Traffic characteristics based on G.711
  codec
• Waiting time in reasonable agreement
• Difference obtained can be attributed to differnt
  queuing strategies

29
Scenario 5 (cont ...)
30
Conclusions

• Queueing model is proposed
• Queueing model caters for the QoS for respective
  service classes
• Results obtained compared with a CSF(OPNET) model
• WiMAX a complex protocol
• Studied the protocol operation
• Evaulated the QoS features for the respective
  service classes
• Effect of Fragmentation,Concatenatio Piggyback
  feature Evaluated
• Effect of contention window size studied ofr BE
  service class
• A Bug is identified in OPNET (DOCSIS)
  implementation

31

• THANK YOU!!!!