Hierarchical Routing

1
Hierarchical Routing

• Via Cluster-Based Control Structures
• Presented by Raquel Harris
• 9-04-02

2
Contents

• Introduction
• Cluster Based Control Structures
• Transmission Management
• Link-Cluster Architecture
• Backbone Formation Clustering
• Route Efficiency Clustering

3
Introduction

• Mobile Wireless Networks are dynamic
• State Changes unpredictable
• Control functions that manage the networks
  performance have to meet these conflicting goals
• Respond rapidly and correctly
• Minimize consumption of networks transmission,
  processing, storage

4
Introduction

• Control Functions
• Performed with respect to a control structure -
  superimposed on the physical network which
  consists of a set of controllers
• For fixed networks the setup of the control
  structure is an administrative task that is part
  of the network design process
• For mobile wireless networks, a self-organizing
  control structure is required

5
Introduction

• Selection of a network control structure and
  algorithm is based on,
• Network size
• Control functions required
• Frequency and magnitude of network state changes
• QOS required

6
Cluster-Based Control Structures

• Cluster-based control structures
• More efficient use of resources in controlling
  dynamic networks
• Physical network is encapsulated by a virtual
  network of interconnected clusters
• Clusters contain one or more controllers which
  make control decisions and may distribute cluster
  state information to outside clusters

7
Cluster-Based Control Structures

• Control Structures attempt to be more efficient
  by,
• Managing transmission to reduce channel
  contention
• Forming routing backbones to reduce network
  diameter
• Abstracting network state information to reduce
  its quantity and variability

8
Transmission Management

• Transmission interference
• Thesis Reduce interference by separation of
  transmission times, space, and frequency or
  spreading code
• Thesis Reduction by coordination of separation

9
Link-Cluster Architecture

• Reduces interference in a multiple-access
  broadcast environment
• Forms distinct clusters for which transmission
  are scheduled in a contention-free manner
• Spread-spectrum multiple access can be used with
  different codes in each cluster
• Automatic organization
• Nodes autonomously organize into interconnected
  clusters
• Union of members is the entire network of nodes

10
Link-Cluster Architecture

• Structure
• Clusterhead
• Each cluster has one or more clusterheads
• Schedules transmissions
• Allocates resources within the cluster
• Gateway
• Each cluster has one or more gateways
• Attempt to have one per pair
• Connect adjacent clusters
• Direct connection member of both clusters
• In-direct connection connected to an adjacent
  gateway

11
Link-Cluster Architecture

• Ordinary Nodes
• Each cluster has zero or more ordinary nodes
• Neither clusterheads or gateways
• Low-delay paths between cluster members
• All members within one hop of the clusterhead
  (within two hops of each other)
• Therefore members of different clusters are at
  least 2 hops away

12
Link-Clustered Architecture
13
Link-Cluster Architecture

• Establishing a link-clustered Control Structure
  involves,
• Discovering neighbors with a bi-directional link
  by broadcasting
• Electing clusterheads and forming clusters
• Agreeing on gateways between clusters

14
Link-Cluster Architecture

• Clusterhead Election
• Identifier-based or Connectivity-based Clustering
• Centralized implementation
• Node with the lowest or highest numbered
  identifier or largest number of neighbors becomes
  the clusterhead
• Distributed implementation
• Node with lowest or highest numbered identifier
  elects itself or its neighbor with the lowest or
  highest id unless that neighbor has relinquished
  clusterhead status
• Node with the highest connectivity of all
  uncovered neighbors becomes clusterhead
• Ties are broken by identifier number

15
Link-Cluster Architecture

• Gateway Election
• Any node with links to more than one cluster is a
  candidate gateway
• If a node has two clusterheads as neighbors, it
  is a candidate gateway connecting two overlapping
  clusters
• If a node has one clusterhead neighbor and can
  reach a second clusterhead in two hops, it is a
  candidate linked to a candidate gateway in
  another cluster

16
Link-Cluster Architecture

• Overlapping Clusters
• Gateway selected is the one with the highest or
  lowest id
• Disjoint Clusters
• Linked pair with one member having the highest or
  lowest id among all candidates connecting the two
  clusters
• Unambiguous selection may require advertisement
  of node ids beyond one-hop neighbors

17
Link-Cluster Architecture

• Node Mobility
• Clusters must be updated accordingly to ensure
  proper scheduling of transmissions
• All nodes must be capable of executing
  clusterhead and gateway functions if the network
  is to be maximally available
• With connectivity-based clustering, clusterhead
  status has potential to change more frequently
  than identifier-based clustering
• Nodes may move out of range of neighbors, losing
  its status as the most connected node

18
Link-Cluster Architecture

• Least Cluster Change Algorithm
• Reduces the number of changes in clusterhead
  status required after node movement
• Clusterhead status change only if two
  clusterheads move within range of each other
• One must relinquish the role
• Or, if a non-clusterhead node moves out of range
  of any other node
• Becomes clusterhead of its own cluster

19
Link-Cluster Architecture

• Routing Backbone
• Traffic is constrained to traverse clusterheads
• Can reduce throughput and robustness of network
• Congestion
• Single point of failure
• Many algorithms do not use clusterheads as
  routing control structure of the network

20
Backbone Formation Clustering

• Routing Backbone
• Reduction of route length is more important for
  multi-hop wireless networks than wireline
  networks
• Larger delays generally experienced at each hop
• Can be formed by expanding the transmission range
  of some or all nodes
• Increasing transmission power may increase
  interference

21
Backbone Formation Clustering

• Near-Term Digital Radio Network (NTDR)
• Designed for mobile tactical communications
• Frequent node movement and outages
• Each cluster has a clusterhead that is linked to
  form the routing backbone
• Nodes are within 1 hop of clusterhead
• Intercluster communication is restricted to
  clusterheads only
• Clusters are gateways
• No multihop communications
• One hop neighbors can communicate directly
• Others intracluster communications must traverse
  the clusterhead

22
Backbone Formation Clustering

• Each node keeps track of the bi-directionally
  linked neighbors by periodically broadcasting
  beacons.
• May become clusterhead if needed
• ltMAC addressgtltpartition idgtltclusterhead
  orggtltcluster membersgtltlink quality from each
  membergtltclusterhead transmit power levelgt
• Used to decide whether to become affiliated with
  the clusterhead
• Each cluster communicates on different
  frequencies
• Isolates intercluster and intracluster
  communications

23
Backbone Formation Clustering

• A low transmission range is used for
  non-clusterhead members
• Allow reuse of frequencies for distant clusters
• Reduce interference
• Clusterhead Election
• Beacon-based only
• No clusterheads detected in its vicinity
• Can heal a network partition

24
Backbone Formation Clustering

• Must limit the number of nodes simultaneously
  participating in clusterhead election
• Each node waits a short random time interval
  before retesting the condition and becoming
  clusterhead if necessary
• Each new clusterhead issues beacons proclaiming
  its status immediately
• Elimination
• Node can relinquish role as clusterhead if,
• It will not partition the network in doing so
• All cluster members can join other clusters
• Checks staggered randomly among nodes over time
  to limit the number of simultaneous relinquishes

25
Backbone Formation Clustering

• Cluster Affiliation
• Preferred clusters
• Same organization affiliation
• Signal from the clusterhead is transmitted at low
  power but received at high strength
• Cluster size is relatively small
• Affiliation can be refused by the clusterhead
• Updated cluster membership is distributed to all
  clusterheads after affiliation has taken place
• Notifies previous clusterhead of new affiliation

26
Backbone Formation Clustering

• Alternate affiliation is sought if,
• Clusterhead relinquishes role
• Clusterheads beacons no longer list member
• Clusterheads beacons indicate the quality of the
  link to the clusterhead has become unacceptably
  poor
• Received signal strength from the clusterhead is
  unacceptably low
• Routing
• Clusterheads share responsibility for maintaining
  the routing backbone
• Monitor and exchange information about changes in
  the backbone

27
Backbone Formation Clustering

• Each generates membership information pertaining
  to its cluster and link-state information
  (resistance metric included) pertaining to its
  links to neighboring clusterheads
• Floods information over the backbone
• Computes least-resistance routes based on this
  information (using Dijkstras SPF) maintains
  next hop to use for each destination
• Must notify other clusterheads of state change
  affecting routing in the backbone once detected
• May result in unacceptable network performance
  due to saturation of update messages in a highly
  mobile network

28
Backbone Formation Clustering
29
Backbone Formation Clustering

• Virtual Subnet Architecture
• Several disjoint routing backbones
• Provides fault tolerant connectivity and load
  balancing in multihop mobile wireless networks
• Structure
• Initially, the network is partitioned into
  physical subnets (disjoint clusters)
• Virtual subnets are formed by clustering the
  different physical subnets
• Assumption transmission power is adjustable to
  retain connectivity
• Neighboring subnets are assigned differing
  frequencies to reduce interference

30
Backbone Formation Clustering

• Subnet clusters and frequencies can be computed
  offline or computed by distributed procedures
• A Maximum number of physical, P, and virtual, Q,
  subnets is predefined for the network.
• Nodes are members of only one physical subnet and
  zero or more virtual subnets
• Ideally one subnet but may belong to more than
  one if it communicates frequently with subnet
  members
• Address uniquely identifies both physical and
  virtual affiliations (multiple address for more
  than one virtual subnet)
• Prefix physical subnet
• Suffix virtual subnet

31
Backbone Formation Clustering
32
Backbone Formation Clustering

• Mobility
• If a node moves out of range of its subnet
• Needs to join another physical subnet, which
  typically leads to a new virtual subnet
  affiliation
• Join a subnet whose Q quota has not been filled
• Otherwise, joins physical subnet as a guest
• New address is announced to all members of the
  physical and virtual subnets
• Guest members do not distribute its address to
  any virtual subnet
• Address Discovery
• Source distributes a query within its physical
  subnet
• Address is returned if the destination is
  affiliated with a virtual subnet of one of the
  sources physical subnet members

33
Backbone Formation Clustering

• A query is distributed within the virtual subnet
  if the destination is not found
• Address is returned if destination is found in a
  physical subnet of one of the sources affiliated
  virtual subnets
• This feature will find guest destinations
• Routing
• Multiple paths are available to nodes within the
  virtual subnet architecture
• Increases fault tolerance
• Direct Routing and long path routing exploits
  this feature

34
Backbone Formation Clustering

• Direct Routing
• Routing by source and destination address
• Next hop selection
• Source seeks node, x, that is member of sources
  physical subnet and destinations virtual subnet
  and forwards packet to it.
• If destination is a guest, no x is found and the
  source uses its virtual subnet to forward the
  packet
• If the network is not partitioned and each
  physical subnet contains a member of each virtual
  subnet direct routing works very well
• Partitions and reconfiguration of the network is
  highly likely in highly mobile networks however.

35
Backbone Formation Clustering
36
Backbone Formation Clustering

• Long-Path Routing
• Routes are randomly distributed to balance
  traffic load and assist the network in
  accommodating partitions
• At most QP-1 different routes
• If subnets are partitioned, intermediate nodes
  will have to make random selections to route
  around partitions
• Source randomly selects a virtual subnet (at
  most Q)
• If the source is not a member of the virtual
  subnet selected, the packet is forwarded via node
  u that is a member of both the virtual subnet and
  the sources physical subnet
• U forwards the packet through the virtual subnet
  to a node v located in the destinations physical
  subnet
• V forwards the packet to the destination node

37
Backbone Formation Clustering

• If the source is a member of the selected virtual
  subnet
• Select at random one of the P-1 (at most)
  physical subnets within its virtual subnet (VS)
  and distinct from its own (PS)
• Forward packet via PS to a node, w, in both the
  sources VS and the PS
• Node w forwards the packet within the PS to a
  node z in the destinations VS and finally to the
  destination

38
Backbone Formation Clustering
39
Routing Efficiency Clustering

• In order to achieve the required network
  performance goals,
• Accurate network state information is required by
  control functions
• Controllers must detect network state changes
• Collection and distribution of network state is
  required
• Mobile networks have a higher rate network state
  changes than fixed networks
• Mobility affect interconnectivity and link
  quality
• Wireless networks add the crucial affect of
  limited and volatile resources

40
Routing Efficiency Clustering

• Updates of current state information may consume
  large amounts of resources
• Transmission, storage and processing
• In highly dynamic networks, updates may lag
  behind the current state
• Sensitivity to network changes is a function of
  the particular control function, resources,
  volatility of the network state, and magnitude of
  state change consequences

41
Routing Efficiency Clustering

• Hierarchical Routing
• Structure
• Network of N nodes is organized into a m-level
  hierarchy of nested clusters
• All level-i clusters are disjoint, 0 ? i ? m
• A node is a level-0 cluster grouped into level-1
  clusters, etc.
• A nodes address are relative to its position in
  the hierarchy
• A concatenation of the labels of the level (m-1)
  through level-0 clusters
• Label of a single level-m cluster is omitted from
  a nodes address
• Node z of figure 4.6 has the address, x.y.z with
  u as the implicit prefix

42
Routing Efficiency Clustering
43
Routing Efficiency Clustering

• Cluster Overlap
• All the routing schemes presented (with the
  exclusion of 1) have disjoint clusters at a
  particular level
• Not a requirement
• Cluster overlap can be desirable
• Node remains reachable after moving out of a
  cluster
• Route length can be reduced
• Multiple clustering hierarchies
• If 2 level 1 clusters a1 and b1 overlap, each of
  their level-I ancestral clusters ai and bi must
  also overlap and a1 ? b1 ? ai ? bi for i ? 1

44
Routing Efficiency Clustering

• Overlap can be expensive
• The number of node addressing information must be
  distributed and maintained in the network is
  proportional to the number of addresses per node
• Route selection is complicated by the need to
  determine which cluster to utilize to maximize
  reachability during movement
• Cluster formation and maintenance can be
  complicated by constraints on the amount of
  overlapped permitted

45
Routing Efficiency Clustering

• Granularity of network state
• Abstraction of network state can reduce the
  amount of resources required to maintain the
  network
• Abstraction can also result in loss of
  information and reduce accuracy of control
  decisions
• Routing Schemes
• Cluster representative a network controller
  that generates and distributes routing
  information for a particular cluster
• Hierarchical routing schemes attempt to prevent a
  single point of failure w.r.t cluster information
• Multiple nodes are capable of quickly assuming
  the role

46
Routing Efficiency Clustering

• Quasi-hierarchical routing schemes
• Most schemes fall into this category
• Most are derivatives of the distance-vector
  routing approach
• Each node learns the next node to use in order to
  reach each level-i cluster within its level-(i
  1) cluster
• Strict-hierarchical routing schemes
• Each node learns the next level-i cluster to use
  in order to reach each level-i cluster within its
  level-(i 1) cluster and
• Each node learns which level-i cluster lies on
  the boundary of its level-(i 1) cluster and
  which are directly linked

47
Route Efficiency Clustering

• Example ck is the lowest-level cluster with the
  source and destination nodes
• K 3
• Quasi-hierarchical routing s0 routes the packet
  directly to the boundary of dk-1. The packet is
  then routed to dk-2 boundary and so on.
• Strict-hierarchical routing S0 routes the packet
  directly to the boundary of sk-1 one level at a
  time. The packet is then routed through
  level-(k-2) clusters in dk-1 to reach the
  boundary of dk-2 and so on.

48
Route Efficiency Clustering
49
Route Efficiency Clustering

• Quasi-hierarchical Routing
• Objective of distance vector routing is to
  determine the next hop on the minimum-cost route
  from a node to each level-i cluster within the
  nodes level(i 1) cluster, for 0 ? i ? m
• A cluster representative advertises to nodes
  outside of its cluster an abstracted cost for
  reaching destinations within the cluster
• Closest entry routing cost 0
• Overall best routing cost average of all
  costs from rep to all nodes in cluster

50
Route Efficiency Clustering

• If x and y are direct linked neighbors belonging
  to the same level-(j1) cluster but different
  level j clusters
• when x receives a new cost for level-i cluster,
  c, contained in xs level-(i 1) cluster, from
  neighbor z, for any 0 ? i ? m, x decides if it
  needs to update its forwarding information for c
  and if it needs to distribute a new cost
  advertisement for c to its neighbors
• If c is not an ancestral cluster of x,
• If Wc-stored gt Wc-z Wx-z x updates its
  forwarding info for c and replaces the next hop
  node with z
• The cost advertisement is updated and sent to
  neighbors x determines require it

51
Route Efficiency Clustering

• If c is an ancestral cluster of x, x simply
  determines the set of neighbors to which it will
  distribute the cost advertisement but does not
  update its forwarding information and cost
  advertisement
• Before distributing the cost ad for c to y, x
  must verify,
• i ? j to prevent propagation of detailed
  routing info outside of that cluster
• The next hop in xs forwarding table entry for c
  is not y to prevent formation of routing loops.
• Reduces the amount of routing info received and
  retained by each node from O(N) to O(mCmax) for a
  m-level hierarchical control structure of nested
  clusters
• Cmax max number of level-i clusters, over all
  i, contained within a level-(i 1) cluster

52
Route Efficiency Clustering

• If each level-i cluster contains the same number
  of level-(i 1) clusters, for 0 lt i ? m, the
  amount of forwarding info at each node mN1/m
• If m is large, the route costs may be
  significantly larger than true minimum-cost
  paths.
• Results of analysis of the quasi-hierarchical
  approach are,
• Number of entries in a nodes forwarding table is
  minimized when the number of level-i clusters in
  each level-(i1) cluster equals e and when the
  number of levels in the clustering hierarchy
  equals ln N ? forwarding table contains elnN
  entries

53
Route Efficiency Clustering

• For any arbitrary source and destination, the
  difference in lengths between route produces by
  quasi-hierarchical routing and the true
  minimum-hop path tends to zero as N ? ?, under
  the following ideal assumptions.
• ? All level-i clusters contain the same of
  level-
• (i 1) clusters, for 0 lt i ? m
• ? Each level-i cluster contains the minimum-hop
• between two nodes resident in that cluster
• ? The diameter of any level-i cluster does not
• exceed bnv c, where n of nodes in the
  
• cluster, 0 ? v ? 1 indicates the
  connectivity of
• the cluster, and b, c are positive
  parameters

54
Route Efficiency Clustering

• Some analysis results indicate that, m ? 4,
  substantial reductions in forwarding table size
  is achieved with only a small increase in route
  length over the true minimum. (results are valid
  within the context of the 3 ideal assumptions and
  2 unrealistic assumptions
• ? All links have equal capacity
• ? Traffic rates between any 2 nodes are
  identical
• Minimum-Cost Paths
• Quasi-hierarchical routing cannot guarantee
  near-minimum-cost routes, ex.,
• di1 cluster is the lowest-level ancestral
  cluster of both the s0 and d0
• di is the level-i ancestral cluster of d0 but not
  S0

55
Route Efficiency Clustering

• Quasi-hierarchical routing yields the minimum-hop
  route from di1 to di and the minimum-hop route
  from di to di-1 and so on
• This value is not necessarily the minimum-hop
  path to d0 unless i 1
• It has been shown, that in worst case, the route
  produced by quasi-hierarchical routing can exceed
  the true minimum-hop path by a factor of 2m 1

56
Route Efficiency Clustering

• A modification to the closest entry routing that
  enables any source to obtain the minimum cost
  route to any destination node has been proposed
• Source determines the minimum-cost route to each
  border node of the destinations level-1 cluster
  using closest entry routing
• Source queries each border node for the cost of
  its minimum-cost route and
• Determines the critical border node node for
  which the sum of costs of the minimum-cost routes
  from the source to itself and from itself to the
  destination is the minimum over all border nodes
  of the destination's cluster

57
Route Efficiency Clustering

• The added cost of queries and responses from
  border nodes of a destination cluster is the
  expense and it is recommended for use on virtual
  circuit forwarding networks where the cost can be
  amortized over the network
• In worse case, the of queries required is
  O(Bm), where B is the max of border nodes for
  any cluster and thus may be impractical for
  networks with many levels or border nodes

58
Route Efficiency Clustering

• Link-State Routing
• Nodes generate, distribute, and use hierarchical
  link-state info
• Level-1 nodes flood its link state to all other
  nodes in the cluster in the form of link costs to
  all of its neighbors
• All nodes can generate minimum-cost routes to all
  other nodes within the cluster using Dijkstras
  SPF algorithm
• Each gate (border node) generates minimum-cost
  routes to all other gates (in the cluster) using
  intracluster link state
• Each gate generates link-state (costs over
  virtual links to neighboring gates within the
  same cluster and in adjacent clusters connected
  via direct links)

59
Route Efficiency Clustering

• Gates flood this link-state to all other gates in
  all level-1 clusters
• All gates have enough info to generate
  minimum-cost routes to all other gates
• Gates flood their clusters with the minimum-cost
  to all other clusters
• Nodes can determine next hop on the minimum cost
  route to any destination cluster
• Route may not be the minimum-cost path since
  there is no means for a source to discover the
  minimum-cost routes connecting gates of the
  destinations cluster to the destination node

60
Route Efficiency Clustering
61
Route Efficiency Clustering

• Hybrid Schemes
• SURAN Survivable Adaptive Networks A DARPA
  program (tactical packet radio networks) for
  which the control structures consists of a
  three-level hierarchy of nodes, clusters, and
  superclusters
• One scheme suggested for SURAN is a hybrid of the
  distance-vector and link-state routing schemes
• Each node uses distance-vector to determine the
  next hop on the minimum-cost route
• Destinations outside of its cluster, the
  distance-vector routing info is partially derived
  from link-state exchanged between clusters and
  superclusters
• Border nodes flood interconnectivity information
  in the form of link-state which is used to
  compute minimum-cost routes to any cluster or
  supercluster within its supercluster or network

62
Route Efficiency Clustering
63
Route Efficiency Clustering

• Focal Nodes
• Help to guide nodes in making forwarding
  decisions
• Permit more direct routes when available
• This makes the focal nodes seldom visited
• Regional Node Routing
• Each level-k cluster (k-region) contains one or
  more k-regional nodes (except for the top level,
  m-region does not contain any nodes)
• A node is both a 0-region and 0-regional node
• A k-regional node is also an i-regional node for
  each i-region in which it resides, for 0 ? i ? k
• Each k-regional node is affiliated with one
  (k1)-regional node in its ancestral (k1) region
• ? address expressed as the concatenation of
• pairs of region and regional node labels
  for its
• ancestral regions and finally the nodes
  label

64
Route Efficiency Clustering
65
Route Efficiency Clustering

• A node may affiliate with more than one regional
  nodes
• Each node in a region, k, distributes
  distance-vector routing info throughout the
  nodes (k1)-region but no further
• A source will consult its forwarding table for a
  reachable node with an address with the longest
  match with the destinations address before
  forwarding a packet
• If no match is found, the packet is discarded
• If multiple entries have the longest match, the
  lowest cost route is selected
• If multiple entries have the lowest cost, a route
  amongst these is selected at random

66
Route Efficiency Clustering

• The selected regional nodes, r, address is added
  to the packet to be used for forwarding decisions
• The packet is forwarded to the next hop
• Each of the remaining nodes, x, that receive the
  packet repeats the same steps as the source and
• Each node uses information about r to select a
  forwarding table entry for an appropriate
  regional node r
• If r ? F (set of regional nodes in xs forwarding
  table and exhibit longest address matches with
  lowest cost routes), x selects r r and
  forwards to the next hop
• If r ? F, x selects r as the source selected r
  if the address match between r and the
  destination is at least as long as that of r and
  the destination and the address of r is replaced
  by r in the packet
• Otherwise, the packet is discarded

67
Route Efficiency Clustering

• The route produced is not necessarily the
  minimum-cost route but is comparable to the
  closest entry route
• If a region is partitioned, a packet may be
  traversed through an alternate regional node,
  provided that the node is at least as close as
  the current but unreachable node
• Landmark Routing
• Each level-i landmark, x, consists of all nodes
  that are within a radius of ri(x) node hops from
  x, 0 lt ri(x) ? ri and ri is the max radius
  permitted for the level-i cluster
• Each level-m landmark has a radius that is at
  least as large as the diameter of the network
• Each node is a level-0 landmark
• More than one level-i landmark is also a
  level-(i1) landmark, 0 ? i lt m and for each, y,
  ri1(y) gt ri(y)

68
Route Efficiency Clustering

• Distance-vector routing is used by each node to
  advertise its cost throughout level-j (highest
  level for which x acts as a landmark)
• x includes a hop count of rj(x) to constrain
  propagation members of level-js cluster
• Each node computes its own cost to x and
  decrements the hop count
• If the decremented hop count gt 0, the node
  distributes the advertisement further with new
  hop count and route cost
• A forwarding table is built by each node where
  each entry refers to a landmark and indicates the
  cost of the next-hop node on the minimum cost
  route to that landmark

69
Route Efficiency Clustering

• The landmark hierarchy is the control structure
  comprised of the landmarks and associated
  clusters
• At level 0, any two clusters associated with
  neighboring nodes must overlap (r0(x) gt 0)
• At level-m, all clusters must overlap since each
  level-m cluster covers the entire network
• Address is a concatenation of landmark labels in
  descending order of level. The relationship
  among specified landmarks satisfies the following
  for 1? i ? m,
• The level-i landmark lies within the cluster
  defined by the level-(i-1) landmark ensures
  that the level i landmark knows how to reach the
  level-(i-1) landmark in a destination address
• The level-0 landmark lies within the cluster
  defined by each level-i landmark ensures the
  node can discover it own address

70
Route Efficiency Clustering
71
Route Efficiency Clustering

• The forwarding table is used to find an entry for
  one of the landmarks specified in the destination
  address.
• If no entries are found, the packet is discarded
• Otherwise, if multiple entries are found, the
  entry with the lowest level among the landmarks
  is selected
• See 4.10(a) and 4.10(b)
• This scheme may increase the risk of packets
  indefinitely wondering the network (for landmarks
  that become unreachable)

72
Route Efficiency Clustering

• Strict-hierarchical Routing
• More robust than quasi-hierarchical schemes in
  the presence of changes in network state
• Increase route costs and packet forwarding
  overhead
• Routing info is assembled by a representative of
  a level-i cluster, c
• The cluster cost is computed
• If c lies on the boundary of its level-(i1)
  cluster, directly linked neighbor level-j
  clusters are determined, j ? i 1
• The routing info is distributed by c using either
  the distance-vector or link-state approach to all
  level-i clusters within the level-(i1) cluster
• Using all information gained, c computes the
  minimum cost routes and next hop clusters from c
  to any other level-i cluster within the
  level-(i1) cluster and determines border
  clusters and their links

73
Route Efficiency Clustering

• The representative of c distributes to all nodes
  within c, the route cost, next-hop cluster and
  cluster boundary info
• The forwarding tables are identical to the
  quasi-hierarchical tables in terms of the number
  and type of entries. A couple of differences are
  evident
• The next hop to c stored is always a cluster at
  level j ? i instead of a node (neighbor of x)
• May need to consult up to 2m-1 forwarding table
  entries to determine the next-hop node to a
  destination
• The cost to reach c from x is computed as the
  sum of individual cluster costs over all level-i
  clusters in the route from c to c. The cost to
  reach c for quasi-hierarchical routing is a true
  minimum since it is the sum of individual link
  costs
• A node determines the longest match between the
  destinations address and the cluster entities in
  its forwarding table before forwarding

74
Route Efficiency Clustering

• A Prototypical Scheme
• Provides route stability for dynamically changing
  nodes and links
• Minimum-cost routes and next-hop nodes are
  determined using distance-vector
• A clusterhead responsible for generating and
  distributing routing info to all clusters in the
  supercluster is selected for each cluster (as in
  figure 4.11 above)
• Each superclusterhead is also a clusterhead for a
  component cluster of the supercluster
• Each clusterhead determines the connectivity and
  cost to each neighboring clusterhead (may not be
  within the supercluster)
• The link-state is flooded to all clusterheads in
  its superclusterhead

75
Route Efficiency Clustering

• Costs from one clusterhead to a neighboring one
  can be
• 1 to indicate intercluster connectivity only
• of cluster hops to the neighboring
  super-clusterhead
• of hops to the neighboring clusterhead
• Each clusterhead uses SPF to compute the minimum
  cost route and next-hop cluster to each cluster
  in its supercluster and to each neighboring
  supercluster in the network
• Similarly, each super-clusterhead computes the
  minimum-cost route and next-hop supercluster to
  each supercluster in the network
• Clusterheads and super-clusterheads do not play a
  central role in packet forwarding

76
Route Efficiency Clustering

• Super-clusterheads distribute to all clusterhead
  in the supercluster info on the next-hop
  supercluster to any supercluster in the network.
• This information is combined with its own to
  obtain the next-hop cluster to any supercluster
  in the network and distributes to each node in
  the network
• Hybrid Scheme
• Designed for the DARPA PRNet program (large
  tactical packet radio networks)
• Strict-hierarchical granularity of forwarding
  table entries
• Quasi-hierarchical granularity and propagation
  of routing information

77
Route Efficiency Clustering

• Distance Vector Routing is used to generate the
  minimum-cost route and next hop to each node in
  the level-1 cluster and discover boundary nodes
  and their connectivity to neighbor clusters
• Route cost is expressed as the number of node
  hops with the next hop being a neighbor node in
  the level-1 cluster
• To route at other levels, global routing nodes
  provide routing information but do not play a
  role in packet forwarding (cluster
  representatives)
• Global routing nodes at level-1 determine the
  minimum-cost route to level-1 clusters and next
  hops to global routing nodes (as it pertains to
  level-(i1), 0 lt i lt m)

78
Route Efficiency Clustering
79
Route Efficiency Clustering

• The quasi-hierarchical routing procedure is
  performed with these differences,
• Route cost is in terms of the of level-1
  cluster hops, therefore,
• ? if x updates its cost to a cluster after
  receiving
• a cost advertisement from neighbor, z, x
  sets
• its cost from x to z to be 0
• ? if a global routing node, g, updates its cost
  to
• a cluster based on a cost advertisement
• received indirectly from a global routing
  node
• g, in neighbor level-1 cluster, g sets the
  cost
• from itself to g to 1

80
Route Efficiency Clustering

• The next hop to the destination cluster is a
  global routing node in a neighbor level-1 cluster
  (not a neighbor node nor a cluster)
• The forwarding table entries includes,
• For each destination in the nodes level-1
  cluster contain the cost of the minimum-hop route
  (node hops) and the next-hop node along that
  route
• For each level-i destination cluster in the
  nodes level-(i 1) cluster, 0 lt i lt m, contain
  the minimum-hop route cost (level-1 cluster
  hops), the next-hop global routing node along the
  route, and the next-hop boundary node (level-1
  cluster) toward the global routing node
• This scheme is considered less robust than the
  SURAN scheme for dynamic networks since changes
  in connectivity of level-1 clusters affect all
  nodes in the network not just nodes in the
  ancestral level-2 cluster

81
Route Efficiency Clustering

• Link-State Routing
• Two schemes (strict-hierarchical) designed to
  provide QoS routing for large dynamic networks
• Common Features
• Link-state approach is used to represent and
  distribute routing info at all levels of the
  hierarchy
• Routes are selected to satisfy individual users
  service requests given the limitations of the
  current state of the network
• Forwarding can be either source-directed (route
  included in packet) or virtual-circuit
• For sessions with many packets, virtual-circuit
  forwarding is preferred since the per-packet
  forwarding overhead is small and the cost of
  establishing the circuit can be amortized over
  all session packets