Online Service Management Algorithm for Cellular/WALN Multimedia Networks

1
Online Service Management Algorithmfor
Cellular/WALN Multimedia Networks

• SOFSEM 2007
• Sungwook Kim
• Sogang University
• Department of Computer Science
• Seoul, South Korea

2
Introduction

• Efficient network resource management
• - key to enhance network performance
  QoS
• Next generation networks
• - support heterogeneous multimedia
  services
• Support heterogeneous multimedia data
• - while ensuring QoS for higher priority
  traffic services
• Traffic pattern is difficult to predict
• - online approach is essential
• Adaptive network management
• - while maintaining a well-balanced
  network performance

3
Online Algorithm

• Online algorithm
• - dealing with the online computation
  problem
• Online computation problem
• - based on past events without future
  information
• - make decisions in real time
• Many QoS problems in network management
• - online computation problems
• The online resource management control
  algorithm
• - natural candidate for multimedia network
  operations

4
Traffic Service

• Traffic services
• ? new and handoff call services in cellular
  network
• - give higher priority to handoff
  services
• ? class I (real-time) and class II (non
  real-time) call services
• in multimedia communication networks
• - class I data service Voice
  telephony, Video-phone
• - class II data service E-mail,
  ftp, Data on demand, etc
• give higher priority to
  class I call services

5
Bandwidth Reservation

• The traffic window size can be adjustable.
• ? If CDPclass_I is higher (lower) than
  Pclass_I,
• - traffic window size is increased
  (decreased)
• - in steps equal to unit_time.
• Bandwidth reservation amount is estimated
  dynamically
• - the sum of requested bandwidth by
  class I calls
• during the traffic window

6
(No Transcript)
7
Buffer Management

• Active Queue Management algorithm
• network router is responsible
• - for detecting network congestion
• - for notifying end hosts of congestion
  to adapt their sending rates
• RED and BLUE algorithms
• - avoid global synchronization
• - adjust the packet dropping probability
  in response to congestion
• - pushing most of the complexity and
  state of differentiated services
• to the network edges

8
RED Algorithm (1)

• The RED (Random Early Detection) Algorithm
• - queue length is used as threshold to
  detect network situation
• - try to maintain an average queue length
  under congestion
• Based on recent buffer history
• - drops incoming packets in a random
  probabilistic manner
• - provide a more equitable distribution of
  packet loss
• - improve the utilization of the network
• Major problem
• - heavily depend on the system parameter
  values
• - average queue length is only the index
  for network situation

9
RED Algorithm (2)

• for each incoming packet
• - calculate the average queue length
  (Avg)
• exponential weighted average
• if Avg lt MINh
• - do nothing
• if MINh lt Avg lt MAXh
• - calculate packet dropping probability
  Pa
• - mark packets with probability Pa
• if MAXh lt Avg
• - mark packet

10
Blue Algorithm (1)

• Recently developed simple algorithm
• - retain all the desirable features of RED
  algorithm
• Main indices of network congestion
• - directly on packet loss and current link
  utilization
• Queue overflow and idle event
• - update the packet marking probability
• - learn the correct rate and send back
  congestion notification
• Major problem
• - queue length variation for bursty
  traffic changes
• difficult to control temporal traffic
  fluctuations

11
Blue Algorithm (2)

• For each packet loss
• ? if ((now last_update) gt freeze_time )
• - Pm Pm Di
• - last_update now
• For link idle event
• ? if ((now last_update) gt freeze_time )
• - Pm Pm - Dd
• - last_update now

12
Orange (Online range) Algorithm (1)

• Three parameter values for QoS and congestion
  control
• adaptive decision by online manner
• ? bandwidth range for the reservation
  (RESb)
• ? queue range (Qr)
• ? packet marking probability (Mp)
• Main issue
• - adaptive range adjustment for bandwidth
  and buffer control
• Orange (Online range) control algorithm
• - adaptive online control for service
  differentiation
• - to provide a better effort service for
  class II traffics
• while ensuring QoS for the admission
  controlled class I services

13
Orange (Online range) Algorithm (2)

• Adjusts system parameters
• - in adaptive online fashion
• Bandwidth reservation range (RESb)
• Queue range (Qr)
• - unused reserved bandwidth can be
  temporarily allocated
• for buffered class II service
• - same as the RESb to maximize network
  performance
• Packet marking probability (Mp)
• - decided proportional to the current queue
  length
• - adaptively characterized by threshold
  values

14
Orange (Online range) Algorithm (3)

• If L lt Qr
• - congestion free
• no arriving packets are dropped
• L gt T
• - all arriving class II data packets are
  dropped
• Qr lt L lt T
• - class II data packets can be marked with
  probability
• Packet marking probability Mp
• 
  
• 
  - L current queue length
• 
  - T maximum buffer size

15
Simulation Model

• Consists of 7 clusters, each cluster consists of
  7 micro cells
• In the even traffic situation, new call arrivals
  are Poisson with rate (0-3 calls/s/cell), which
  is uniform in all the cells
• In the uneven traffic situation, the arrival rate
  of hot cell is Poisson with rate 3
• Capacity of each cell is C (30Mbps)
• One base station per cluster is selected randomly
  as the faulty base station and this occurs at a
  random time
• Mobiles can travel in one of 6 directions with
  equal probability with three cases of user
  velocity
• Eight different data groups are assumed based on
  call
• duration, bandwidth requirement and class
  of service
• Durations of calls are exponentially distributed
  with different means for different multimedia
  data types

16
Simulation Results

• Fig.1 Call Blocking Probability
  Fig.2 Call Dropping Probability

17
Concluding Remarks

• Development of efficient bandwidth management
• - for QoS sensitive multimedia networks
• Proposed integrated online approach
• - provides excellent network performance
  while ensuring QoS
• guarantees under widely different
  traffic scenarios
• On-line decisions based on real time estimates
• - mutually dependent each other
• - adaptable and quite flexible to traffic
  changes
• Strike the appropriate balanced network
  performance
• - among contradictory QoS requirements
  while other existing
• schemes cannot offer such an attractive
  trade off

18
.
Internet Communication Control (ICC) Research
Lab.Prof. Sungwook Kim
19
Internet

• Differentiated Services (DiffServ)
• ? Complexity Scalability
• - easy to implement
• - no state information is needed in the
  core routers
• does not suffer from the
  scalability problems
• - concentrates on packet forwarding
• using appropriate queue management
• Major problem
• ? QoS control
• - not to provide guaranteed QoS for
  higher priority traffic services
• growing interest in Internet
  QoS

20
Bandwidth Reservation (1)

• guarantee QoS for class I data traffic services
• ? maintain the reserved bandwidth close to
  the optimal value
• ? on-line estimate by traffic window
• - based on real time measurement
• - keeps the history of class I task
• - learn the pattern of coming requests
• - close to the optimal value
• - partition the time axis into equal
  interval
• unit_time

21
Bandwidth Reservation (2)
22
Bandwidth Reservation (3)

• The traffic window size can be adjustable.
• ? If CBPclass_I is higher (lower) than
  Pclass_I,
• - traffic window size is increased
  (decreased)
• - in steps equal to unit_time.
• Bandwidth reservation amount is estimated
  dynamically
• - the sum of requested bandwidth by
  class I calls
• during the traffic window

23
Online management for Internet

• Guarantee QoS for class I data traffic services
• ? maintain the reserved bandwidth close to
  the optimal value
• ? on-line estimate by traffic window
• - based on real time measurement
• ABlink
• MABpath(i,j)

24
Call Admission Control (1)

• CAC is responsible to decide
• - granted, declined or renegotiated
• Two system parameters are used
• ? One-way packet Delivery Time (ODT)
• packet delay time of setting path
• ? the Acceptance Threshold (AT)
• the predefined bit sending rate
• Network probing
• - to determine if all routers along the
  path have available bandwidth

25
Call Admission Control (2)

• For a new class I request,
• - a probing packet estimates the available
  network bandwidth
• SR bits/sec ( BU
  ) ATi bits/sec
• For a new class II request,
• - a probing packet only estimates the unused
  network bandwidth
• SR bits/sec ( BU
  ) M_ATj bits/sec
• Guarantee QoS for class I data traffic services

26
Internet

• The rapid growth of data communication network
• - Internet Protocol (IP) Internet
• - QoS sensitive multimedia data services
• based on different priority
• Major Problem
• - difficult to support guaranteed QoS
• bounded delay minimum
  throughput
• for higher priority real time
  applications

27
Intserv Model

• Integrated Services (IntServ)
• - in order to provide QoS in Internet.
• - signal to the network through a
  reservation request
• ReSerVation Protocol (RSVP)
• - end-to-end signaling protocol
• - receiver-oriented protocol for setting
  up resource reservations
• - reservations have to be refreshed
  periodically
• Major problem
• ? Complexity Scalability
• - router has to keep state
  information on all reservations

28
Diffserv Model

• Differentiated Services (DiffServ)
• ? Complexity Scalability
• - easy to implement
• - no state information is needed in the
  core routers
• does not suffer from the
  scalability problems
• - concentrates on packet forwarding
• using appropriate queue management
• Major problem
• ? QoS control
• - not to provide guaranteed QoS for
  higher priority traffic services
• growing interest in Internet
  QoS

29
AQM Algorithms

• Active Queue Management algorithm
• network router is responsible
• - for detecting network congestion
• - for notifying end hosts of congestion
  to adapt their sending rates
• RED and BLUE algorithms
• - avoid global synchronization
• - adjust the packet dropping probability
  in response to congestion
• - pushing most of the complexity and
  state of differentiated services
• to the network edges

30
RED Algorithm (1)

• The RED (Random Early Detection) Algorithm
• - queue length is used as threshold to
  detect network situation
• - try to maintain an average queue length
  under congestion
• Based on recent buffer history
• - drops incoming packets in a random
  probabilistic manner
• - provide a more equitable distribution of
  packet loss
• - improve the utilization of the network
• Major problem
• - heavily depend on the system parameter
  values
• - average queue length is only the index
  for network situation

31
RED Algorithm (2)

• for each incoming packet
• - calculate the average queue length (Avg)
• exponential weighted average
• if Avg lt MINh
• - do nothing
• if MINh lt Avg lt MAXh
• - calculate packet dropping probability Pa
• - mark packets with probability Pa
• if MAXh lt Avg
• - mark packet

32
BLUE Algorithm (1)

• Recently developed simple algorithm
• - retain all the desirable features of RED
  algorithm
• Main indices of network congestion
• - directly on packet loss and current link
  utilization
• Queue overflow and idle event
• - update the packet marking probability
• - learn the correct rate and send back
  congestion notification
• Major problem
• - queue length variation for bursty
  traffic changes
• difficult to control temporal traffic
  fluctuations

33
BLUE Algorithm (2)

• For each packet loss
• ? if ((now last_update) gt freeze_time )
• - Pm Pm Di
• - last_update now
• For link idle event
• ? if ((now last_update) gt freeze_time )
• - Pm Pm - Dd
• - last_update now

34
Online Control in Internet

• Basic idea of the cellular network management
• - can be applied to Internet
• Online strategy based on real time measurements
• - due to the uncertain network environment
• do not require advance knowledge or
  prediction
• Major advantage of an online approach
• - adaptability, flexibility,
  responsiveness to current traffic conditions
• Online algorithm based on DiffServ model
• - provides QoS guarantees for higher
  priority calls
• while accommodating as many call
  connections as possible

35
Multimedia Internet Management

• Online management algorithm
• ? the QoS provisioning mechanism
• - guarantee QoS based on call admission
  control
• for class I data service
• ? the congestion control mechanism
• - adaptive bandwidth allocation for
  higher network performance
• for class II data services
• Integrated online approach
• - both mechanisms act cooperatively
• in order to simultaneously
  satisfy the conflicting requirements

36
Orange (Online range) Algorithm

• Three parameter values for QoS and congestion
  control
• adaptive decision by online manner
• ? bandwidth range for the reservation
  (RESb)
• ? queue range (Qr)
• ? packet marking probability (Mp)
• Main issue
• - adaptive range adjustment for bandwidth
  and buffer control
• Orange (Online range) control algorithm
• - adaptive online control for service
  differentiation
• - to provide a better effort service for
  class II traffics
• while ensuring QoS for the admission
  controlled class I services

37
Online Control Algorithm for Internet

• QoS guarantee for higher priority service
• - no reduction in network capacity
• Ability to adaptively congestion control
• - to maximize network performance
• Low complexity
• - practical for real network
  implementation
• Ability to respond to current network traffic
  conditions
• - for the appropriate performance
  balance
• between contradictory QoS
  requirements

38
QoS provisioning mechanism (1)

• During network congestion
• - QoS provisioning problem is further
  intensified
• Admission control management
• - provide good QoS in Internet
• Link bandwidth is shared dynamically
• - between class I and class II data
  services
• - each service has different operational
  requirements
• Different admission control rules
• - strict admission control rule for class
  I data services
• - non-controlled admission rule for class
  II data services

39
QoS provisioning mechanism (2)

• Bandwidth is partitioned by range
• - some part is reserved for higher priority
  traffic service
• - partition range can be movable
• Bandwidth range (RESb) for reservation
• - adaptive adjustment by traffic window
• ? online computational problem
• Admission decisions for class I traffic services
• - controlled by the moving range
• get the benefit from reservations
  for QoS guarantees

40
Congestion control mechanism (1)

• On-line control for network congestion
• unable to optimally control the network
  congestion exactly
• ? try to close to optimal network
  performance
• - responsive to current traffic
  changes in link loads
• - adaptive balance between traffic
  history
• 
  and recent traffic changes
• Dropping packet rate
• - provide feedback information
• the congestion level of the
  gateways through the path

41
Congestion control mechanism (2)

• Adjusts system parameters
• - in adaptive online fashion
• Bandwidth reservation range (RESb)
• Queue range (Qr)
• - unused reserved bandwidth can be
  temporarily allocated
• for buffered class II service
• - same as the RESb to maximize network
  performance
• Packet marking probability (Mp)
• - decided proportional to the current queue
  length
• - adaptively characterized by threshold
  values

42
Congestion control mechanism (3)

• If L lt Qr
• - congestion free
• no arriving packets are dropped
• L gt T
• - all arriving class II data packets are
  dropped
• Qr lt L lt T
• - class II data packets can be marked with
  probability
• Packet marking probability Mp
• 
  
• 
  - L current queue length
• 
  - T maximum buffer size

43
Congestion control mechanism (4)

• Recent traffic patterns reflect effectively the
  current condition
• - during recent unit_time tc -
  unit_time, tc
• Traffic management in next interval
• - adaptively control packets during tc,
  tc unit_time
• L lt Qr
• - packet queuing rate (Ip_r) in current
  interval
• packet incoming rate - packet
  clearing rate
• ? if (T Qr ) lt Ip_r then Mp1
• Qr lt L lt T
• ? if (0 lt Ip_r ) then Mp2
• ? if (Ip_r lt 0) Ip_r gt (L Qr)
  then no packet drop

44
Online Control Practical Applications

• Dynamic QoS priority control in multimedia
  networks
• - call priority can be changed based on
  online requests and
• current network conditions
• Main concept of this dissertation
• ? integrated online approach based on
  real-time measurement
• - develop other adaptive control
  algorithms
• - inter-process communication, disk and
  memory
• file and I/O systems, CPU scheduling,
  power control,
• distributed operating system

45
Concluding Remarks

• QoS guarantee for higher priority service
• - no reduction in network capacity
• Ability to adaptively congestion control
• - to maximize network performance
• Low complexity
• - practical for real network
  implementation
• Ability to respond to current network traffic
  conditions
• - for the appropriate performance
  balance
• between contradictory QoS
  requirements