Load Balanced Link Reversal Routing in Mobile Wireless Ad Hoc - - PowerPoint PPT Presentation

Load Balanced Link Reversal Routing in Mobile Wireless Ad Hoc Networks

Nabhendra Bisnik, Alhussein Abouzeid ECSE Department RPI Costas Busch CSCI Department RPI

Mobile Wireless Networks

 Wireless nodes are mostly battery driven ) limited transmission range  Nodes act as relays  Often involves many-to-one communication

 Multihop wireless mesh networks  Mobile sensor networks

 Link reversal routing (LRR) is a good choice

 Loop free routes  Low overhead

 However LRR may lead to unbalanced distribution

• f load (traffic forwarded)

Contributions

 Identify the causes of load unbalance in LRR Propose three heuristic mechanisms that attack different causes of load unbalance  Evaluate the performance of the heuristics using simulations

Talk Outline

 Link Reversal Routing  Causes of load unbalance  Load balancing problem  Heuristic mechanisms  Simulations

Talk Outline

 Link Reversal Routing  Causes of load unbalance  Load balancing problem  Heuristic mechanisms  Simulations

Link Reversal Routing

 Properties

 Distributed  Loop free at every instant  Low overhead  Offers both proactive and reactive modes  Multiple routes to destination

 Two phases

 Route creation phase  Route maintenance phase

Route Creation Phase

Destination Height = 0 1 2 2 3 4 5 1 1 1 1 1 1 QRY UPD 1 2 2 3 4 5 Directed Acyclic Graph (DAG)  Route creation phase assigns height to each node and transforms connected network into a DAG  a ! b exists in the DAG only iff h(a) > h(b)  Thus DAG is loop free  In general h(a) = [h1(a), h0(a) ] where h1(a) = height assigned by LRR and h0(a) = node id of a  Lexicographical ordering used

Route Maintenance Phase

1 2 2 3 4 5 1 2 2 3 4 5 3 2 2 3 4 5 3 6 4 3 4 5 7 6 4 7 4 5 7 6 8 7 8 5 Full Link Reversal Algorithm  Brings network from a bad state to a good state  Runs in (n2) time  Leads to increase in height of at least one node

Talk Outline

 Link Reversal Routing  Causes of load unbalance  Load balancing problem  Heuristic mechanisms  Simulations

Causes of Load Unbalance

 Traffic flows from higher height to lower height  Each time a node looses route to the destination, its height increases  The nodes with stable routes to destination tend to have lower height  Thus stable nodes relay large amount of traffic leading to

 Battery exhaustion  Congestion

Load Unbalance - Example

7 6 8 7 8 5 7 6 8 7 8 5 Although alternate path is now available, most of the traffic is still routed through the node with height 5

Unbalanced Network State

 If there exist routes to the destination in the undirected network graph whose use may lead to a more uniform spread of load, but the routes are absent in DAG  Characteristics of unbalanced network state

 Selfish nodes (nodes with no incoming links)  High height gradients (h(a) – h(b) > 2 and a ! b exists in the DAG)

C B A D H E F I G J K

1 2 1 2 2 3 4 3 4 5 Selfish Node Isolated Routing Components

L 6

High Height Gradient

Talk Outline

 Link Reversal Routing  Causes of load unbalance  Load balancing problem  Heuristic mechanisms  Simulations

Load Balancing Problem

 Two Components of the problem

 Maintaining a good DAG ()  Use good forwarding strategy over the DAG (S)

 Forwarding Strategy maps a link l of the DAG to traffic flowing over it, xS(l)  Total traffic forwarded by a node where E(i) is the set of outgoing links of node i  Load balance metrics

 Balance Factor (BF)  Squared Sum (SS)

Load Balancing Problem

 From optimization point of view, the load balance problem is to find and s.t.  This problem is NP-hard, distributed solution is even more difficult Or,

Talk Outline

 Link Reversal Routing  Causes of load unbalance  Load balancing problem  Heuristic mechanisms  Simulations

Heuristic Mechanisms

 Three heuristic mechanisms

 Selfish Node Based Mechanism (SNBM)  Proactive Decrease in Height (PDH)  Reactive Increase in Height (RIH)

 Height manipulation

 Decrease height ) attract traffic  Increase height ) repel traffic

Selfish Node Based Mechanism

 Aims to balance the size of isolated routing components  Periodically each node checks if it is selfish  If node selfish then

 If hmax – hmin > 2 then

 Sets height to minimum height that ensures path to the destination  Fix link directions  Update neighors

Selfish Node Based Mechanism

C B A D I E F J H K L

2 7 1 2 2 3 4 3 6 8

G

5

C B A D I E F J H K L

2 2 1 2 2 3 4 3 4 5

G

5

C B A D I E F J H K L

2 7 1 2 2 3 4 3 6 5

G

5

C B A D I E F J H K L

2 2 1 2 2 3 4 3 4 3

G

5

C B A D I E F J H K L

2 2 1 2 2 3 4 3 4 3

G

5

Selfish Node Based Mechanism

 However every instances of load unbalance does not involve selfish nodes  Example )  Solution – reduce height whenever it is possible in

• rder to balance DAG

 This observation leads to PDH

C B A D I E F J H K L

2 1 2 2 3 4 3 6 8

G

5

M 9

7

Proactive Decrease in Height

 Each node periodically compares its height with neighbors  If it is possible to decrease height without becoming a sink, then

 Set height to minimum possible height that allows route to destination  Fix link directions  Update neighbors

Proactive Decrease in Height

C B A D I E F J H K L

2 1 2 2 3 4 3 6 8

G

5

M 9

7

C B A D I E F J H K L

1 1 2 2 3 4 3 6 8

G

5

M 9

7

C B A D I E F J H K L

1 1 2 2 3 4 3 4 8

G

5

M 9

2

C B A D I E F J H K L

1 1 2 2 3 4 3 4 3

G

5

M 4

2

Reactive Increase in Height

 Both SNBM and PDH are proactive in nature  RIH acts only when needed  Each node records the amount of traffic forwarded during an update window  If load served during an update window exceeds threshold then

 Set height equal to hmax + 1  Fix link directions  Update neighbors

Reactive Increase in Height

A B C D G F E

5 4 3 6 5 4

A B C D G F E

5 4 5 6 5 4

A B C D G F E

5 6 5 6 5 6

A B C D G F E

5 6 5 6 7 6

A B C D G F E

5 6 5 8 7 6

Forwarding Strategies

 Load distribution is also affected by the forwarding strategies  Two forwarding strategies considered

 Multi-path routing

 Distribute load equally among all downstream links  Requires maintenance of forwarding records

 Shortest path routing

 Forward packets to downstream neighbor that lies on the shortest path available in the DAG  Requires no state information

Talk Outline

 Link Reversal Routing  Causes of load unbalance  Load balancing problem  Heuristic mechanisms  Simulations

Simulation Setting

 N mobile nodes, initially deployed randomly

• ver 1000m £ 1000m area

 Communication radius is m  Random waypoint mobility model used with vmin = 2m/s, vmax = 5m/s, pause time = 5s  Each node generates traffic at rate 1Kbps, destined to a sink node  Sink node located at (500m, 500m)  Models mobile wireless sensor network, multi-hop wireless mesh networks

Performance Metrics

 Balance factor and squared sum for both multi-path and shortest path forwarding  Network lifetime  Routing updates

Balance Factor

 PDH has highest balance factor  As number of nodes increases, path length increases leading to lower balance factor Multi-path routing has larger balance factor Multi-path routing Shortest path routing

Squared Sum

Multi-path routing Shortest path routing  Again PDH has smaller squared sum  Multi-path routing leads to longer routes, hence larger squared sum

Network Lifetime

 PDH leads to highest network lifetime  Lifetime decreases with increase in number of nodes

Height Update Rate

 An update message is produced each time height of a node is updated  Thus routing overhead is proportional to the height update rate  RIH may cause a chain reaction of height updates, thus has much higher

• verhead

Conclusion and Future Work

 All the proposed schemes achieve better load balance than basic LRR  PDH is the best, since it is most aggressive  Future Work

 NS-2 implementation of the proposed schemes  Approximate algorithms based on optimization framework

Thank You!