Distributed QOS Routing in Ad Hoc Networks

1
Distributed QOS Routing inAd Hoc Networks

• Shigang Chen
• Klara Nahrstedt

2
Agenda

• Introduction
• System Models
• Algorithm Overview
• Delay Constrained Routing
• Bandwidth Constrained Routing
• Dynamic Path Maintenance
• Simulations and results

3
Introduction - 1

• QOS Routing
• Wireless environment
• Imprecise state information.
• Data Traffic Best effort Vs QOS
• gt It is difficult to provide QOS in Dynamic
  Environment.

4
Introduction - 2

• The proposed Algorithm
• Distributed QOS Routing scheme for Ad Hoc Nets
• Delay Constrained
• BW Constrained
• Properties
• Imprecise state information.
• Multipath parallel routing.
• Consideration of general cost
• Fault tolerance techniques for paths maintenance.

5
Introduction - 3

• Network DynamicsThe algorithm works well if the
  average life time is much longer than average
  rerouting time. (Order of magnitude longer.)

6
Agenda

• Introduction
• System Models
• Algorithm Overview
• Delay Constrained Routing
• Bandwidth Constrained Routing
• Dynamic Path Maintenance
• Simulations and results

7
System Models

• Ad Hoc Network Model
• Stationary and transient Links
• QOS State Matrix
• The Routing Problem
• Imprecise State Model

8
System Models Ad Hoc Network Model - 1

• V nodes interconnected by E full duplex
  communication links.
• V and E are changing over time.
• Each node has at least one transmitter and one
  receiver.
• Effective transmission distance for all
  transmitters is equal.
• Two nodes have link between them and are
  neighbors if they are in transmission range of
  each other.

9
System Models Ad Hoc Network Model - 2

• Every node i knows the set of its neighbors Vi
• We assume there is a MAC protocol
• resolves media connection
• Supports resource reservation
• Keeps only messages intended for hime.g. J.
  tsai, M. Gerla Multicluster, mobile, multimedia
  radio network ACM-Baltzer J. Wireless Networks
  vol1, no3.

10
System ModelsStationary and transient Links

• Transient links
• links between nodes that move fast.
• Exists only for a short time.
• Stationary links
• links between nodes that move slowly.
• Assuming continuous connection
• preferable
• Vis j (i,j) is a stationary link, j ? Vi
• All nodes in Vis are stationary.
• All nodes in Vi - Vis are transient.
• Imprecise classification.

11
System ModelsQOS State Matrix - 1

• A node i has a precise information about its
  neighbors
• delay (i,j).
• bandwidth (i,j).
• cost (i,j).(transient link cost much more)

12
System ModelsQOS State Matrix - 2

• Path P i -gt j -gt -gt k -gt l.
• Delay (P) delay(i,j)delay(k,l).
• Bandwidth(P) minbandwidth(i,j),,bandwidth(
  k,l).
• Cost (P) cost(i,j) cost(k,l).

13
System ModelsThe routing Problem - 1

• Delay Constrained Routing Problem
• S Source node.
• t - Destination node.
• D Delay requirement.
• Find a feasible path P from S to t such that
  delay(P) D.When there are multiple such paths,
  select the one with least cost.

14
System ModelsThe routing Problem - 2

• Bandwidth Constrained Routing Problem
• S Source node.
• t - Destination node.
• B Bandwidth requirement.
• Find a feasible path P from S to t such that
  bandwidth(P) B.When there are multiple such
  paths, select the one with least cost.

15
System Models Imprecise State Model - 1

• Every node i keeps state information for every
  destination t
• Di(t) The delay of the least delay path from i
  to t.
• Bi(t) The bandwidth of the largest bandwidth
  path from i to t.
• Ci(t) The cost of the least cost path from i to
  t.
• ?Di(t) Delay variation. (see explanation in
  next slide)
• ?Bi(t) - Bandwidth variation. (see explanation
  in next slide)
• Di(t), Bi(t) , Ci(t) are inherently imprecise.

16
System ModelsImprecise State Model - 2

• ?Di(t) Delay variation is the estimated maximum
  change of Di(t) before the next update.(The
  actual end to end delay is expected to be between
  Di(t) - ?Di(t) and Di(t) ?Di(t) )
• ?Bi(t) - Bandwidth variation is the estimated
  maximum change of Bi(t) before the next
  update.(The actual end to end bandwidth is
  expected to be between Bi(t) - ?Bi(t) and Bi(t)
  ?Bi(t) )

17
System ModelsImprecise State Model - 2

• The algorithm doesnt apply the imprecision model
  for Ci(t) in sake of simplicity.
• Example of computing ?Di(t)
• At each state update (MAC) keep
• Diold(t) , Dinew(t), ?Diold(t) , ?Dinew(t)
• Let d be a factor to determine how fast to forget
  the history. (0lt dlt1)
• ?Dinew(t) d ?Diold(t) (1- d) Dinew(t)-
  Diold(t).

18
System ModelsImprecise State Model - 3

• The imprecision model does not intended to cover
  every possible situation, but only to improve
  the average performance.

19
Agenda

• Introduction
• System Models
• Algorithm Overview
• Delay Constrained Routing
• Bandwidth Constrained Routing
• Dynamic Path Maintenance
• Simulations and results

20
Algorithm Overview

• Ticket-based probing A multipath distributed
  routing scheme
• Limited number of search paths No flooding.
• Intelligent hop by hop path selection.
• Probes (routing messages) are sent to search for
  a low cost path that satisfies the QOS
  requirements.

21
Algorithm Overview

• The goalTo collectively utilize the state
  information at the intermediate nodes to guide
  the routing messages (probes) along the best
  paths to the destination, so that the probability
  of finding feasible (satisfies QOS req.)
  low-cost path is maximized.

22
Algorithm Overview

• A source node issues a number of tickets based on
  the available state information.
• Each routing message (probe) contains at least
  one ticket.
• A probe can be split by intermediate node if it
  contains more then one ticket.
• The total number of probes (and search paths) is
  bounded by the number of tickets issued by the
  source node.
• An Intermediate node decides whether an incoming
  probe should be split and to which of its
  neighbors to forward the probe.

23
Algorithm Overview
24
Algorithm Overview

• Properties
• Routing overhead dynamically controlled.
• Imprecise information- When the state information
  is highly imprecise we will use more tickets.
• Distributed routing.
• Local and E2E information at intermediate nodes
  is used Can correct some of the wrong decision
  made by the source.

25
Agenda

• Introduction
• System Models
• Algorithm Overview
• Delay Constrained Routing
• Bandwidth Constrained Routing
• Dynamic Path Maintenance
• Simulations and results

26
Delay Constrained Routing

• Introduction
• Determining the Number Of Tickets
• Forwarding The Received Tickets
• Termination and path selection
• Data Structure
• Rerouting
• Soft States
• Local Multicast

27
Delay Constrained RoutingIntroduction - 1

• Heuristic algorithm
• Parameters
• S source node.
• t destination node.
• N0 Number of tickets initially issued by the
  source.
• N(p) Number of tickets in a probe p.
• Each probe accumulates the delay of the path it
  traversed so far.
• If the accumulated delay gt D (QOS constrain) ,
  the probe will be terminated.

28
Delay Constrained RoutingIntroduction - 2

• Two problems
• How the source determine N0 ?
• How an intermediate node distributes the arrived
  tickets among the outgoing probes.
• Guidelines (Problem 1)
• For connections with large delay requirements one
  ticket will be issued.
• For connections with very small delay required
  No ticket will be issued.
• For the rest of connections smaller delay
  required -gt more tickets issued.

29
Delay Constrained RoutingIntroduction - 3

• Guidelines (Problem 2)
• Node i received N(p) tickets.
• Tickets will be distributed unevenly depends on
  their chances of leading to reliable low-cost
  feasible path
• A neighbor having lower E2E cost to t should
  receive more tickets than neighbor with larger
  E2E cost to t.
• A neighbor having lower E2E delay to t should
  receive more tickets than neighbor with large E2E
  delay to t.
• A neighbor having stationary link to i should be
  given more tickets then a neighbor with transient
  link to i.

30
Delay Constrained RoutingDetermining the Number
Of Tickets - 1

• The N0 tickets are colored with yellow and green
• Y0 yellow tickets probes prefer paths with
  smaller delay.
• G0 green tickets probes prefers paths with
  smaller cost.
• Overall strategyUse Green tickets more
  aggressively to maximize the chance of finding
  feasible low-cost path.Use the yellow tickets as
  a backup to guarantee a high success probability
  to find a feasible path.
• N0 Y0 G0.

31
Delay Constrained RoutingDetermining the Number
Of Tickets - 2

• Number of yellow Tickets
• Based on D The delay requirement.
• Large Delay Only one yellow ticket is issued.
• Too small Delay No yellow ticket is issued and
  the connection is rejected.
• Otherwise several tickets are issued based on the
  previous guideline.
• Number of green Tickets
• Based on D The delay requirement.
• Large Delay Only one green ticket is issued.
• Too small Delay No green ticket is issued and
  the connection is rejected.
• Otherwise several tickets are issued based on the
  previous guideline.
• Parameters
• F The maximum number of yellow tickets.
• T Threshold coefficient.
• O The maximum number of green tickets.

32
Delay Constrained Routing Determining the Number
Of Tickets - 2

• Determining the number of yellow tickets

33
Delay Constrained Routing Determining the Number
Of Tickets - 3

• Determining the number of green tickets

34
Delay Constrained Routing Forwarding the
received Tickets - 1

• Reminder
• Path of probe P i -gt j -gt -gt k -gt l.
• delay (P) delay(i,j)delay(k,l).
• Probes proceed only if delay(P) D.
• Every node i keeps data about all its neighbors
  j
• Dj(t) The estimated delay between node j and
  node t.
• ?Dj(t) The delay variation.These parameters
  can be enquired via distance vector alg.

35
Delay Constrained Routing Forwarding the
received Tickets - 2

• Suppose Node i received probe P from node k with
  Y(P) yellow tickets, G(P) green tickets.
• Rip Neighbors of nodes i which are candidate
  for forwarding the probe.Rip j delay(P)
  delay(i,j) Dj(t) - ?Dj(t) D ,j?Vis - k
  .Vis j (i,j) is a stationary link, j ? Vi
• If Rip 0 then Rip will be recalculated for all
  neighbors of i, not just the stationaryRip
  j delay(P) delay(i,j) Dj(t) - ?Dj(t) D
  ,j?Vi - k .

36
Delay Constrained Routing Forwarding the
received Tickets - 3

• Distributing Y(P) , G(P) over Rip
• For every candidate j, node i make a copy of P
  denotes as Pj .
• ?j?Rip G(Pj) G(P).
• ?j?Rip Y(Pj) Y(P).
• A probe sent into a path with less expected delay
  should get more yellow ticketsY(Pj) Y(P) /
  delay(i,j) Dj(t) ?j?Rip
  delay(i,j)Dj(t)-1
• A probe sent into a path with less expected cost
  should get more green ticketsG(Pj) G(P) /
  cost(i,j) Cj(t) ?j?Rip
  cost(i,j)Cj(t)-1
• If G(Pj) Y(Pj) lt 0 then Pj is sent to node j.

37
Delay Constrained RoutingForwarding the received
Tickets - 4

• In order to prevent loops
• Only one probe may be sent on each outgoing link
  for each connection request.
• The number of hops a probe may traverse can be
  bounded.
• Probes record their path. If loop occurs the
  probe is discarded.

38
Delay Constrained RoutingTermination and path
selection

• Every node Upon which Rip 0 Invalidates the
  tickets and sends them to the destination t.
• The routing process terminates when all tickets
  have arrived to t or time out elapsed.
• If several valid paths were found, the
  destination t selects the least cost path as the
  primary and the rest will be secondary paths that
  can be used if the primary is broken.
• A confirmation message is sent to the source and
  initiates resource reservation on the reverse
  path.
• If the resource reservation fails or the reverse
  message fails the destination will be notified
  and it will select a new path from the secondary
  paths.

39
Delay Constrained RoutingData structure of a
probe

• id system wide unique identifier for the
  connection request.
• s source node.
• t destination node.
• D - delay requirement.
• Y0 G0 Total number of tickets.
• k sender of P.
• Path the path P which the probe has traversed
  so far.
• Y(P) Number of yellow tickets carried by P.
• G(P) Number of green tickets carried by P.
• delay(P) accumulated delay of the path so far.
• cost(P) accumulated cost of the path so far.

40
Delay Constrained RoutingRerouting

• Rerouting
• Nodes join, move and exit the network.
• The load on links change along time
• Rerouting can be done periodically and/or
  triggered when a broken path is detected.

41
Delay Constrained RoutingSoft States

• Routing and rerouting can be done in conjunction
  with RSVP based on soft states.
• Soft state
• Resource reservation must be refreshed
  periodically
• If a soft state is not refreshed with in a
  timeout it is deleted.
• A refreshing message is sent from the destination
  to the source back along the reversed routing
  path periodically.
• When an intermediate node receives the refreshing
  messages it resets its time out counter.

42
Delay Constrained RoutingLocal Multicast

• The broadcast transmission mechanism of Ad Hoc
  nets can be used to transmit a single local
  multicast message containing the tickets of all
  neighbors.

43
Agenda

• Introduction
• System Models
• Algorithm Overview
• Delay Constrained Routing
• Bandwidth Constrained Routing
• Dynamic Path Maintenance
• Simulations and results

44
Bandwidth Constrained Routing

• Bandwidth Constrained routing share the same
  principles as Delay Constrained routing.
• Modifications are made to the Data structures,
  ticket creation curves and ticket distribution
  formulas.
• Bandwidth delay routing will not be discussed
  here.

45
Agenda

• Introduction
• System Models
• Algorithm Overview
• Delay Constrained Routing
• Bandwidth Constrained Routing
• Dynamic Path Maintenance
• Simulations and results

46
Dynamic Path Maintenance

• Reminder The algorithm works well if the average
  lifetime is much longer than average rerouting
  time. (Order of magnitude longer.)
• In the following section we will discuss
• Detection of a broken path.
• Rerouting.
• Path redundancy.

47
Dynamic Path MaintenanceDetection of a broken
path.

• s source
• t destination.
• i intermediate node.
• Suci any successive node to node I on path P
• Every node I uses a neighbor discovery protocol
  
• If node i discovers that neighbor Suci no longer
  exists, all connections using link (Suci ,i) are
  declared broken.

48
Dynamic Path MaintenanceRerouting

• When node i discovered that link (Suci ,i) is
  broken, it searches the source s for all
  connections using this link and sends a
  path-breaking message.
• The source of each connection initiates a new
  ticket based probing protocol and sends a release
  message to the resources.
• During rerouting data packets are sent on a best
  effort traffic.

49
Dynamic Path MaintenancePath redundancy - 1

• The ticket-based routing can find multiple
  feasible paths.
• Successive runs of the algorithm with some
  modifications can generate disjoined paths (if it
  is feasible).
• First Level Redundancy
• Establish several paths, prefer disjoined paths.
• Every data message will be transmitted to all
  available paths.
• When the number of paths will be less then some
  minimum, a new ticket based algorithm will be
  activated to find more paths.

50
Dynamic Path MaintenancePath redundancy - 2

• Second Level redundancy.
• Find several paths.
• Reserve resources for all the paths.
• Data packets will be transmitted only by the
  primary path.
• When the primary path is detected broken, a
  secondary path will be activated.
• When the number of paths will be less then some
  minimum, a new ticket based algorithm will be
  activated to find more paths.

51
Dynamic Path MaintenancePath redundancy - 3

• Third Level redundancy.
• Find several paths.
• Dont Reserve resources for the secondary paths.
• Data packets will be transmitted only by the
  primary path.
• When the primary path is detected broken, a
  message will be sent on the secondary path to
  check if resources still available.

52
Dynamic Path MaintenancePath repairing

• Every node i keeps its successors successor
  (Suci2 ).
• When a path (Suci ,i) detected broken, node i
  will broadcast a repair_message(Suci2, i).
• All nodes in the broadcast range will transmit
  back to i their available resources.
• Node i will select the best new path according to
  the guidelines denoted at slide 29 and transmit a
  reroute message to the source s.
• The source s will check if the E2E delay is
  violated.

53
Simulations and results

• Success Ration Number of connections accepted /
  total number of connection requests.
• Av. Msg. overhead total number of messages
  sent/ total number of connection requests.
• Av. Path cost total cost of all established
  paths / number of established paths.
• Mobility ratio total moving time / (total
  moving time total station time)
• Qos tatio total Qos time / (total Qos time
  total best effort time).

54
Simulations and results -1
55
Simulations and results -2
56
Simulations and results -3
57
Simulations and results - 4
58
Simulations and results - 5
59
Simulations and results - 6
60
Simulations and results - 7
61
Simulations and results - 8
62
Simulations and results 9
63
Simulations and results - 10
64
Simulations and results - 11
65
Simulations and results - 12
66
Simulations and results - 13
67
Simulations and results - 14
68
Simulations and results - 15
69
Simulations and results - 16
70
Simulations and results - 17
71
Simulations and results - 18