Algorithms for Routing and Centralized Scheduling to Provide QoS in IEEE 802.16 Mesh Networks

1
Algorithms for Routing and Centralized Scheduling
to Provide QoS in IEEE 802.16 Mesh Networks

• Presented by Hermes Y.H. Liu

2
Authors

• Harish Shetiya and Vinod Sharma
• Dept of Electrical Communication Engineering,
  Indian Institute of Science
• ACM conference WmuNeP05, Oct 13, 2005 Montreal,
  Quebec, Canada

3
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

4
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

5
Introduction

• 802.16d Mesh mode
• Stationary, do not support mobility
• Centralized scheduling scheme
• Mesh Base Station (MBS) a node that has a direct
  connection to backhaul services outside the Mesh
  network
• Mesh Subscriber Station (MSS)
• Does not specify an algorithm for scheduling nor
  any routing algorithm

6
Introduction

• This article provide algorithms for centralized
  scheduling of real and non real time traffic with
  QoS in 802.16 mesh mode
• 1.Fix the routing within the network, develop
  scheduling algorithms provide QoS to real time
  voice and video (UDP)
• 2.Develop algorithms provide QoS to interactive
  data uses TCP
• 3.Combine both real and non real time
  applications
• 4.Discuss admission control for sufficient
  resource concern

7
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

8
System Model

• Two physical layers
• WirelessMAN-OFDM in the licensed band and
  WirelessHUMAN in the unlicensed band, both use
  256 point FFT OFDM TDMA/TDM for channel access
• Supports only Time Division Duplex (TDD) to share
  the channel between uplink and downlink
• MBS periodically collects channel information and
  requests of all nodes to make centralized
  scheduling

9
System Model

• M MSSs labeled 1,2,.....M, MBS is labeled 0
• is the data rate and is the average
  data rate of the channel from node i to node j
• Real time application IP telephony and Video
  conferencing use UDP
• Non real time application Interactive data (ex.
  Web browsing) use TCP

10
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

11
Routing

• Have the same route for all the traffic at a node
• Develop algorithm has good performance for UDP
  and TCP even it is not optimal for either
• The route should be fixed
• 1.time varying routing can cause loops and may
  require resequencing at the receiver which cause
  performance degradation
• 2.To provide QoS guarantees, resources will be
  reserved along the route. This is possible only
  if the route is fixed

12
Routing

• Assumptions
• 1.The channel states stay same during a frame
• 2.From frame to frame they change independently
  forming an i.i.d. sequence (stationary and
  ergodic)
• 3.The packets arriving in frame k at a node can
  be serviced in the next frame only
• 4.The external arrivals to each node form an
  i.i.d. sequence
• 5.Each node has an infinite buffer to store the
  packets

13
Routing
Assigned transmission rate
External arrivals
Arrivals from other nodes to nodes i for output link (i,j) during frame k
queue length at node i for output link (i.j) in the beginning of frame k

time slots assigned in link (i,j)
14
Routing

• Where denotes max (0,x)
• For the queue to be stable, we need

Where is under the stationary
distribution
15
Routing

• Let
• 
  
• Where are
  the nodes whose data passes through node i
• The entire system is stable if
  for all i1...M
• Since (total time slots), we
  get
• where are the
  nodes through which the data of node i is routed
  

16
Routing

• For each node i, if we choose the route that
  minimizes
• , the overall stability region can be maximized
  and also minimizes the average transmission time
  to transmit a packet from a node to the MBS
• This route can be found by standard shortest path
  algorithms (Dijkstras or Bellman-Ford) by
  assigning cost to link(i,j)

17
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

18
QoS for Real Time Traffic

• QoS in real time UDP connections like IP
  telephony and video conferencing
• 1.End to end delay of a packet should not exceed
  150 msec
• 2.If a packet exceeds this delay, it will be
  dropped
• 3.The dropped probability should be less than 2
• We proposed that at the end of a frame we drop
  the packets which could not be transmitted
  through the wireless network
• Audio encoders (IP telephony) has a constant bit
  rate (CBR)
• Video encoder (MPEG) has a variable bit rate (VBR)

19
QoS for Real Time Traffic

• 4.1 Scheduling of CBR traffic
• The scheduling problem for CBR-UDP is to
  calculate the number of slots
  required at node , such that
  units of data can be transmitted to the MBR and
  the end to end drop probability is bounded by

Constant, the amount of traffic generated by users at node i during a frame
The upper bound of the drop probability at node i
Nodes through which the data of node i traverses
20
Scheduling of CBR Traffic
At node the drop probability is bounded by



The number of slots required, has to
satisfy
This reduces to
Rewrite it as
is the pdf of the link rate which is
assumed to be known
21
Scheduling of CBR Traffic

• Consider the optimization problem
• Subject to
• And
• Can calculate the scheduling and find the number
  of slots

22
Scheduling of VBR Traffic
Be the amount of data generated by flow j (VBR
flows) in frame k

• Assume the arrival process
  for each J is
• stationary and ergodic with known statistics
• Also assume the arrival process from various
  sources are
• mutually independent
• Find such that

• is called the equivalent bandwidth of
  the VBR source, the MBS
• can treat VBR as a CBR flow generating
  units of data per frame and calculate the number
  of slots required

• In practice, the exact statistics of a VBR
  arrival process may not be
• available, we calculate the value of by
  using a source model has
• all the known characteristics but has the worst
  case behavior

23
Simulations

• Consider a 10 nodes Mesh network (Fig. 1)

24
Simulations

• Physical layer characteristics (Tab. 1)
• Burst profiles and data rates (Tab. 2)

25
Simulations

• Frame duration is 10 ms, scheduling is done over
  3 frames
• Scenario 3 CBR flows and 10 VBR flows at each
  node
• Desired rate CBR 64 kbps, an upper bound of
  60ms and 2 on the delay and drop probability
• VBR source is characterized by a 4 states Markov
  chain with the rates 20kbps, 40kbps, 60kbps,
  80kbps. The transition matrix is
• Each flow requires a delay bound of 60ms and a
  drop probability bound of 2
• The equivalent bandwidth of the VBR is 66.9 kbps

26
Simulations

• All the flows get their desired QoS

27
Simulations

• The average bandwidth provided is greater than
  even the maximum
• Required bandwidth

28
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

29
QoS for TCP Traffic

• Emails do not require any QoS, but web traffic
  and file transfer may require certain minimum
  throughput
• In addition to ensuring the minimum throughout
  requested, we need to allocate excess bandwidth
  fairly to different TCP connections
• QoS to two different TCPs
• Persistent TCP long lived connections need to
  send a large file, QoS requirement is the minimum
  mean throughput
• TCP-ON-OFF A TCP connection transfers multiple
  files. Between transfer of two files, a TCP
  connection may not have a file to transfer (OFF
  period) for sometime. QoS requirement is the mean
  file download time

30
QoS for TCP Traffic

• Let be the minimum throughput of all TCP
  connections at node i
• Fairness issues will be considered
• When the route is fixed via the shortest path
  routing, we compute the total number of slots to
  be allocated (Fixed Allocation Scheme) and use
  the channel conditions to adapt it (Adaptive
  Fixed Allocation Scheme)

31
Fixed Allocation Scheme

• Notation
• Allocate a fixed number of slots per frame to
  each node depending upon The average data arrival
  rate and the estimate average channel rate

Minimum throughput (in bytes per frame) required by TCP traffic generated at node i
Mean rate of link i
Number of slots to be allocated at node j
32
Fixed Allocation Scheme
Constraint
(C1)
(C2, end to end throughput is is the same)
(C3, proportional fairness)
,i 1,...M
Combining (C1), (C2), (C3), With , it is a
proportionally fair allocation provide the
maximum throughput to different nodes
33
Fixed Allocation Scheme

• Let . This is the number of
  slots allocated by the node j for the total TCP
  traffic passing through it
• will not ensure the TCP traffic generated at
  node i will get its share of slots at node
  j
• For TCP traffic originating at node i, we form a
  separate queue at each node of its route. Via WRR
  (Weighted Round Robin) out of the total
  allocation of slots, we provide slots to
  this queue at node j
• Problem
• 1.Links can be in bad state or
• 2.nodes do not have enough data to transmit at a
  given time.
• In Adaptive Fixed Allocation Scheme use the
  instantaneous channel state and queue length
  information to adapt

34
Adaptive Fixed Allocation Scheme

• If , we declare the link
  to be bad
• where is the link rate and
  is the predefined threshold rate
• Let G, B denote the set of good and bad links
• If , then ,
  else

If
If
Then the number of slots to node i in frame k in
the first round is
If
otherwise
where is the smallest integer
greater than x
35
Adaptive Fixed Allocation Scheme

• The first round , the
  remaining are allotted in the second
  round
• Second round preference
• 1.nodes with good links, positive credits and
  MAX data
• 2.good links with MAX data
• 3.bad links with MAX data
• The allocation is done one slot at a time,
  recalculating the credits and remaining data to
  transmit after each slot
• Credit update

36
Adaptive Fixed Allocation Scheme

• In second round keep the same order of preference
  without the MAX data transmission condition which
  is so called Channel Adaptive Fixed Allocation
  Scheme
• We can optimize the parameters (
  ) to utilize some system performance (e.g.
  maximizing overall TCP throughput in the system)
• Once we get the slot allocation for each node j
  via fix or adaptive algorithms, we provide the
  slots to node i via WRR

37
Simulations

• In Mesh network Fig. 1
• 12 TCP persistent flows originating at each node,
  6 in uplink and 6 in downlink
• 3 classes of traffic (throughput requirement
  40kbps, 80kbps, 120kbps) with 2 flows each at a
  node
• 10 TCP-ON-OFF flows originating at each node, all
  in uplink
• 2 classes of TCP-ON-OFF flows with 5 flows in
  each class
• Mean packet size is 1000 bytes
• Delays in the external network are arbitrarily
  fixed 0, 60

Class 1 T(on)3s T(off)4s D25 packets
Class 2 T(on)3s T(off)5s D50 packets
38
Simulations
(Percentage difference)
All the fixed schemes satisfy QoS of most of the
users. The Adaptive fixed allocation scheme
provide more than the min required throughput to
most of the flows. So more flows can be supported
with this scheme
39
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

40
Joint Scheduling of UDP and TCP Flows

• UDP? consider the worst case channel condition
• TCP? consider the average channel rate
• The difference between average bandwidth
  requested and the average bandwidth provided will
  be utilized for TCP flows
• UDP has higher priority than TCP. In that case,
  the delays experienced by UDP can be drastically
  reduced without affecting the throughput of TCP
  flows. Also it will save resources

41
Joint Scheduling of UDP and TCP Flows

• Let denote the total number of slots to
  node i to meet the QoS requirements of UDP flows
• Let be the average throughput required by
  all UDP
• Let be the total throughput required by TCP
  flows
• To provide a mean throughput ,the
  can be calculated by the Fixed Allocation Scheme
• Ensure QoS to all UDP connections while TCP get
  their average throughput
• In Channel Adaptive, considering the channel
  conditions we can defer the allocation of the
  other slots

42
Simulations

• Mesh network in Fig. 1 and characteristics
• 3 CBR, 3VBR, (UDP) 12 TCP (6 uplink, 6 downlink)

43
Simulations
Compare to Fig. 2 the available resources are
almost fully utilized
44
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

45
Admission Control

• The number of slots allotted for node i at node j
  is
• Suppose now a new TCP connection request for a
  bandwidth of arrives. MBS applies Fixed
  Allocation Scheme first.
• The number of slots required at the node for
  the new request is
• The connection is accepted if
• If the above condition is not met, then the
  Adaptive Allocation Scheme is used
• If there is no additional resource after these
  algorithms, the new request is rejected

46
Admission Control

• The arrival of a new UDP connection, if VBR
  traffic, the MSS determines the class to which it
  belongs, recomputes the effective bandwidth and
  MBS calculates the number of slots in methods
  above.
• If the required slots is less than the number of
  slots available then the connection is accepted
  otherwise not

47
Outline

• Introduction
• System Model
• Routing
• QoS for Real Time Traffic
• QoS for TCP Traffic
• Joint Scheduling of UDP and TCP Flows
• Admission Control
• Conclusion

48
Conclusion

• Designed algorithms for routing and centralized
  scheduling in IEEE 802.16 mesh networks.
• Considered end to end QoS
• Handled UDP TCP traffic separately and considered
  them jointly
• Provided admission control policy

49
Thanks for your listening

• Hermes Y.H. Liu