Research Article Discovering the Influences of Complex...

Research ArticleDiscovering the Influences of Complex Network Effects onRecovering Large Scale Multiagent Systems

Yang Xu Pengfei Liu Xiang Li and Wei Ren

School of Computer Science and Engineering University of Electronic Science and Technology of ChinaChengdu Sichuan 611731 China

Correspondence should be addressed to Yang Xu xuyanguestceducn

Received 26 December 2013 Accepted 6 March 2014 Published 2 April 2014

Academic Editors H R Karimi X Yang and W Zhang

Copyright copy 2014 Yang Xu et alThis is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Building efficient distributed coordination algorithms is critical for the large scale multiagent system design and the communica-tion network has been shown as a key factor to influence system performance even under the same coordination protocol Althoughmany distributed algorithm designs have been proved to be feasible to build their functions in the large scale multiagent systems asclaimed the performances may not be stable if the multiagent networks were organized with different complex network topologiesFor example if the network was recovered from the broken links or disfunction nodes the network topology might have beenshifted Therefore their influences on the overall multiagent system performance are unknown In this paper we have made aninitial effort to find how a standard network recovery policy MPLS algorithm may change the network topology of the multiagentsystem in terms of network congestion We have established that when the multiagent system is organized as different networktopologies according to different complex network attributes the network shifts in different ways Those interesting discoveriesare helpful to predict how complex network attributes influence on system performance and in turn are useful for new algorithmdesigns that make a good use of those attributes

1 Introduction

In state of the art of artificial intelligence research scalablemultiagent system applications in complex environmentshave been popular in domains of military [1] crisis manage-ment [2] and business [3] In those systems to efficientlycoordinate and share information agents are required tocommunicate via flexible wireless media such as mobilead hoc networks In those networks agents may only beable to directly connect with a few of the others and thenetwork topologies may dynamically change according toagentsrsquo movements or joint intentions Moreover when thesystem gets bigger Musolesi et al have found that the systempresents the characteristics of complex networks [4] forexample small world effect discovered by Travers and Mil-gram [5] and scale free phenomenon discovered by Barabasiand Albert [6] Predicting team performances according todifferent communication network topologies is interestingand challenging From our previous research we learned thatthe same coordination algorithm may lead to huge different

performances when the complex network topologies vary [7]Other researches support our discovery as well For exampleScerri and Sycara mathematically analyzed how differentcomplex networks may affect the system in informationsharing sensor fusion and task allocation [8] Gaston andDesJardins took a bottom-up research on network formationand found that the incomplete complex network structurevaries in decentralized adaptation strategies on team perfor-mances [9]

Although the effects of complex networks are popular inlarge multiagent systems not all distributed algorithms aretested under different complex network topologies in orderto estimate how complex network effects change the systemperformance As the network recovery is one of the mostimportant network operations to maintain a desired systemperformance in this paper we made an initial effort on find-ing how the network topology of a multiagent system shiftswhen agents recover from their communication failures Forexample a UAV may be shot by a hostile missile or robotsrsquoconnections broken by physical obstacles in a mobile ad

Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 407639 13 pageshttpdxdoiorg1011552014407639

2 The Scientific World Journal

hoc network Although the popular restoration of reroutingmechanism for exampleMPLS recovery algorithm has beenproven to be feasible on network recoveries the potentialshifts on the multiagent network topology may significantlychange the system performance in an unpredictable way

In our simulation we simulated a series of large scalemultiagent systems with different complex network topolo-gies A popular restoration of rerouting mechanism-MPLSalgorithm is implemented to recover link and node failuresThe network is evaluated by its diameter average distanceand cluster [10] To simulate the physical network in thoseexperiments the communication capabilities of each link ornode are limitedThe experiment results are presented in twomajor sections system robustness and influence of networktopologies

In the first section of the experiments data that flowsthrough the failure agents or links has to be rerouted andmay cause congestions on existing links or agents Basedon our discussion about the efficiency of network recoverythe number of newly congested links or nodes (agents) thathurt system performance is investigated More importantlythe network is in danger of being disconnected from thosecongestions By comparing the influence on different networktopologies while there are some differences in different casesour major discoveries are that a random network is morelikely to be congested if more links are broken while asmall world network is the least likely On the other handa scale free network appears to be more vulnerable to nodecongestions while a grid network is the most robust

In the second section we test how the network topologymay be shifted from the network recovery We have foundthat most of network topologies are immune to the networkrecovery when the congestions are not so serious Howeverwhen the number of congested links and nodes is highlyincreased the network topologies may have been changedIt is especially the case when the network is organized as ascale free network or a small world network However sincescale free and small world networks are the most importantattributes in a large multiagent system from our previousresearch experience their changesmay significantly affect thesystem performances

2 Related Work

The structure and behavior of complex networks have beenattractive in various studies [11] Inspired by the discoverieson how the rich network structure facilitates effective orga-nizational behavior [12] Gaston and DesJardins illustratedthe importance of network topologies in multiagents net-works [9] Y C Jiang and J C Jiang analyzed the complexnetwork in actor-oriented and actor-structure views to findthe relationship between complex networks and multiagentsystems [13] Taylor et al examined joint actions in themultiagent optimization problem and the results are sur-prising because a high number of connections in a complexnetworked multiagent team can hurt system performanceeven communication and computation costs are ignored [14]Liu et al proposed an integrated model based on small

Figure 1 Different complex network topologies random networkgrid network small world network and scale free network

world network and multiagent system to simulate epidemicspatiotemporal transmissionThey found that the small worldeffect brings better performance than the traditional modeland his discovery has been applied in real geographicalmultiagent applications [15]

On the other hand a set of researches show that net-work operations could influence the performance of thelarge scale multiagent systems as well For example whenfacing the same network attack the network vulnerabili-ties are different if the multiagent systems are organizedas various types of complex networks [16] DrsquoAngelo andFerretti explained that gossip protocols would change thecommunication of the complex network and have someimpacts on routing efficiency [17] Gong andXu analyzed andtested that different parameters on a scale free network canmake significant different efficiencies of information delayin multiagent systems [18] Peschlow et al [19] described aflexible dynamic partitioning algorithm that rapidly recoversinformation routing and optimizes the performance withdifferent complex network topologies

3 Modeling the Complex Networks

The network topology of a multiagent system is defined asan undirected graph 119866 = (119881 119864) where 119881 = 1 2 119873119873 = |119881| defines the agent set 119864 denotes the set of linksbetween agents that is link (119894 119895) isin 119864 and 119894 and 119895 areneighbors 119899 119881 rarr 119864 defines the set of all neighbors of anagent That is 119899(119894) = 119887 119895 119897 is the neighbor of agent 119894 119866could be organized as different network topologies based onthe different properties of complex networks In this paperwe are mainly interested in four of them shown in Figure 1random network grid network small world network andscale free network Preliminary studies [20] have found thateach topology encodes different fundamental properties thatis network diameter average distance between nodes clusterand degree distributions

(i) Degree the degree of agent 119894 is 119889(119894) = |119899(119894)|

(ii) Average degree 119889 = (1119873)sum119894isin119881

|119899(119894)| is the averagenumber of neighbors of all agents for any complexnetwork 119889 ≪ |V|

(iii) Degree distribution 119901(119896) = 119875119903[119889 = 119896] defined as afraction of agents (the number of such agents is 119889)with the degree 119896

(iv) Distance distance(119894 119895) is defined as the least numberof hops to communicate between the agents 119894 and 119895Specifically distance(119894 119895) = 1 if (119894 119895) isin 119864

The Scientific World Journal 3

(v) Average distance

119897 =1

119873 (119873 + 1)sum

forall119894119895isin119881

distance (119894 119895) (1)

is the average distance between any pairs of agents(vi) Network diameter the diameter of the graph 119866 is

defined as argmax(119863) where 119863 = distance(119894 119895) |

119894 119895 isin 119881 which is the longest distance between pairsof agents

(vii) Clustering coefficient 119862119894for an agent 119894 is given by

the proportion of links between the agents withinthe set of 119899(119894) that is divided by the number of linksbetween them The set 119890(119894) is to record the existinglinks between these agents in 119899(119894) then

119862119894=

2 |119890 (119894)|

|119899 (119894)| lowast (|119899 (119894)| minus 1) (2)

The clustering coefficient for the network is given by119862 = (1119873)sum

119873

119894=1119862119894

Different complex network topologies can be describedaccording to the properties mentioned above Erdos andRenyi put forward a classical random network ERmodel [21]In this model a random network follows a Poisson degreedistribution Most nodes in a grid network keep the samedegree which is also called a regular network Watts andStrogatz put forward the concept of small world networkand the WS model [22] This model presents much shorteraverage distance than that in a grid network Moreover sometypical large scale networks such as mobile agents on internet[23] and hyperlinks on web [24] possess certain dynamicsmdashMatthew effect [25] a power law distribution 119901(119896) prop 119896

minus119903

(2 lt 119903 lt 3) Some researchers found an interesting formulaargmax(119863) prop lnln119873 [26] that the average distance maydecrease as the network grows [27] This formula preciselyreflects the small world effect as well Then Barabasi andAlbert put forward a scale free network and theBAmodel [6]Our simulations are based on those complex networkmodels

To simulate a physical network we define the flow of thecommunication for a link (119894 119895) isin 119864 as 119891(119894 119895) and its allowedbandwidth is set as a constant written as 119865max Therefore119891(119894 119895) should not overflow to its bandwidth otherwise thelink will be congested Please note if 119891(119894 119895) = 0 thereis no communication in a physically connected link (119894 119895)

and it is called a backup link that may be used for futurecommunications In addition we define 119903(119894) as the amountof communication through agent 119894 119903(119894) cannot be more than119862max(119894) the max allowed bandwidth capability through agent119894 otherwise the agent is congested as well Moreover thefollowing properties are defined in our simulations

(i) Network connection 119866 is disconnected if exist119894 119895 isin 119881distance(119894 119895) = infin or no value can be assignedOtherwise we say the network 119866 is connected

(ii) Subgraph of a pair of agents ⟨119894 119895⟩ let 1198661015840(119894 119895) be thesubgraph of 119866 1198661015840(119894 119895) = (119881

1015840 1198641015840) where 119881

1015840 is the setof agents on all the shortest paths between the pairs

(1) find all transition agents 119878(119894)(2) for each agent 119908 isin 119904(119894) do

(3) 119878119890119899119889119863119886119905119886(119908119891 (119894 119895)

|119904 (119894)|)

(4) end for

Algorithm 1 119871119894119899119896 119877119890119888119900V119890119903119910((119894 119895) 119891(119894 119895))

(1) 119901119886119905ℎ(119894) larr 119864119899119906119898119890119903119886119905119890119878119905119903119890119886119898(119894)(2) for each path 119901 isin 119901119886119905ℎ(119894) do(3) (119906 V) larr 119865119894119899119889119873119890119894119892119887119900119903119904(119901 119894)(4) 119871119894119899119896 119877119890119888119900V119890119903119910((119906 V) 119891(119901))(5) end for

Algorithm 2 119860119892119890119899119905 119877119890119888119900V119890119903119910(119894)

of agents ⟨119894 119895⟩ and 1198641015840sub 119864 consists of all the links in

those shortest paths(iii) Transition agents of agent 119894 to 119895 let 119904(119894) be the set to

record all the neighbors that can transfer data fromagent 119894 to 119895 and 119904(119894) sub (119899(119894) cap 119866

1015840)

(iv) Transition paths of agent 119894 let path(119894) =

⟨119904 119905⟩ | 119904 119905 isin 119899(119894) be the setthat records all the pairs of agents that communicatedata via agent 119894

4 MPLS-Based Recovery Mechanisms

When agents come to link failures or node failures therestoration of rerouting mechanisms is usually exploited tomaintain the communication between agents In this paperwe implement a typical network recovery policy MPLS[28] to restore communication between agents by reroutingmechanisms

Algorithm 1 briefly describes how a failed link (119894 119895)whosedata flow is 119891(119894 119895) is recovered We suppose there is apredefined sequence according to agent1015840 ID that 119894 ≺ 119895Assume that each agent is able to get the global state of thenetwork agent 119894 can easily find all the alternative shortestpaths to 119895 and 119908 is one of the transition agents 119904(119894) (line1) In line 3 the data flow will be divided evenly into piecesaccording to the number of shortest paths detected Each ofthem will be sent to one of the transition agents and passedthrough predefined paths (line 3)

Algorithm 2 briefly shows how a failed agent 119894 is recov-ered The communication through an agent 119894 may be com-posed of several streams from different links Each stream119901 going through the agent 119894 is written as a unique path 119906 119894 V where 119906 ≺ V The value of data flow goingthrough 119901 is written as 119891(119901) Therefore to recover nodefailure Algorithm 2 first enumerates all the stream path(119894)(line 1) For each stream 119901 we will find the pair of neighborsof 119894 ⟨119906 V⟩ in 119901rsquos path (line 3) Then if we suppose there hasbeen a link of (119906 V) whose communication amount is 119891(119901)


(1) 120583 larr 119908(2) for each agent 119894 isin 119904(119894) do(3) if agent 119894 was not been visited then(4) 119909 larr 119865max minus (120583 + 119891(119894 119895))(5) if 119909 lt 0 then(6) 120576 larr 119865max minus 119891(119894 119895)(7) return the link (119894 119895) is congested(8) end if(9) if 120583 = 0 then(10) 119905 larr 119862max minus (120576 + 119903(119895))(11) end if(12) if 119905 le 0 then(13) return the agent 119895 is congested(14) else(15) if 119905 le 120576 then(16) 120583 larr 119905(17) return the agent 119895 is congested(18) else(19) 120583 larr 120576(20) end if(21) end if(22) end if(23) end for

Algorithm 3 119862119900119899119892119890119904119905119894119900119899(119894 119895)

the stream 119901 can be rerouted in a way similar to link failurealgorithm (Algorithm 1) (line 4)

5 Network Robustness after MPLS Recovery

AlthoughMPLS recoverymechanism can effectively enhancerecovery efficiency network congestion cannot be avoideddue to the limited physical communication capacities ofthe links and agents In order to detect network conges-tion including link congestions and node congestions wedesigned Algorithm 3 to check existed network congestionsaround the multiagent system by rerouting data from a linkfailure (119894 119895) or a node failure 119895

According to Algorithm 3 we can first summarize thelink congestion by link (119894 119895) It is supposed that the amountof messages conveyed to the link (119894 119895) is set as 120583 (line 1) Theavailable capacity 119909 for an existing link is calculated as 119909 =

119865maxminus(120583+119891(119894 119895)) (line 4) If there is no available capacity link(119894 119895) is congested (lines 5ndash7) To judge the agentsrsquo congestionwe calculate the available capacity 119905 for an existing agent 119895

as 120576 larr 119865max minus 119891(119894 119895) and 119905 larr 119862max minus (120576 + 119903(119894)) (lines 6ndash10) If there is no available capacity the agent 119895 is congested(lines 12ndash21) The rerouting mechanisms modify the LSP in afailed spot and the length of the shortest path is often morethan one agent therefore the amount of messages 120583 may bemodified (lines 16ndash19) and the algorithm may be recursivelyjudged many times in terms of Depth First Search (DFS)

In this section we investigate how network recoveringoperation may create node or link congestions when 119866 isorganized as four different network topologies The systemsize is 119873 = 1000 average degree is 119889 = 6 and maximum

allowed link overload is 119865max = 10 The agent having ahigher degree usually plays an important role in the networktherefore the communication through the agent is largerBased on this observation the capacity of agent 119894 follows119862max(119894) = 120582 times 119889(119894) times 119865max where 0 lt 120582 lt 1 is a constant sothat its capability is proportional to the degree of the agentThe results are evaluated according to newly congested linkscongested agents probability of network connectivity and theaverage distance of the network by varying the ratio of failedlinks (Ratio link) the ratio of failed nodes (Ratio agent)and the ratio of average communication of each link 119891(119894 119895)

(Ratio flow) Moreover when119898 = 0 we assume there are nobackup links and all the links are used for communicationand for each (119894 119895) 119891(119894 119895) gt 0 If 119898 gt 0 additional 15backup links are presented The amount of communicationfor each link is randomly setThe experiment results are basedon 100 runs

51 Robustness on Link Failure Recovery In this section weexamine the number of congested links and congested nodeswhen the MPLS algorithm recovered link failures We fixthe average communication of each link 119891(119894 119895) as 50 ofthe 119865max (119877119886119905119894119900 119891119897119900119908 = 05) In Figures 2(a) and 2(b)we varied the number of link failures from 05 to 5of all the links In Figure 2(a) network recovery operationleads to congested links and congested nodes and theirnumber rapidly increases with the slow increase ofRatio linkBy comparing the number of congested links in the sameRatio link we have found that random network appears tohave the largest number of congested links while the smallworld network appears to have the least number of congestedlinks


Ratio link

300

250

200

150

100

50

00005 0010 0015 0020 0025 0030 0035 0040 0045 0050

Num

ber o

f con

geste

d lin

ks

Random networkScale-free networkGrid networkSmall-world network

(a) Link congestion (vary percentage of failed link)

Ratio link0005 0010 0015 0020 0025 0030 0035 0040 0045 0050


0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des

(b) Node congestion (vary percentage of failed link)

Ratio flow

Num

ber o

f con

geste

d lin

ks


01 02 03 04 05 06 07 08 09

0

50

100

150

200

250

(c) Link congestion (vary average flow)

Ratio flow01 02 03 04 05 06 07 08 09


0

20

40

60

80

100

120

140

Num

ber o

f con

geste

d no

des

(d) Node congestion (vary average flow)

Figure 2 Congestions from link failure recovery by varying Ratio link and Ratio flow

The experimental results in Figure 2(b) show that exceptfor the grid network network recovery operation on failedlinks will lead to node congestions Apparently scale freenetwork appears to have the largest number of congestedagents in any settings because of the stability of hub nodeswhich have large bandwidth (proportion to its degree) Theother agents with limited bandwidth are prone to be jammedMoreover cluster may contribute to the node congestionsas well For example the agent 119894 whose degree is largerthan othersrsquo in a scale free network has a larger clusteringcoefficient 119862

119894 and the number of agents retransmitting data

would be high In addition the cluster of a grid network is the

smallest in the four networks and the node congestions areless likely to be present

In the next experiment we fix the failed link as 2(119877119886119905119894119900 119897119894119899119896 = 002) Figures 2(c) and 2(d) show that whenthe average communication Ratio flow of each link 119891(119894 119895)

increases from 10 to 90 of the 119865max both the congestednodes and links increase nomatterwhat the complex networktopologies are Consistent with the conclusion from Figures2(a) and 2(b) a random network performs the worst whilea small world network performs the best according tocongested links A scale free network appears to have thelargest number of congested nodes and random network



Ratio link0005 0010 0015 0020 0025 0030 0035

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

00

(a) Connectivity (vary percentage of failed link)


Ratio link00050000 0010 0015 0020 0025 0030 0035

0

5

10

70

80

90

100

Aver

age d

istan

ce

(b) Average distance (vary percentage of failed link)

Ratio flow

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05 06 07


(c) Connectivity (vary average flow)


Ratio flow

0

5

10

70

80

90

100Av

erag

e dist

ance

00 01 02 03 04 05 06 07

(d) Average distance (vary average flow)

Figure 3 Connectivity and average distance by varying Ratio link and Ratio flow

performs worse as well The grid network appears to have theleast number of congested nodes It can be explained the sameas Figure 2(b)

Figure 4 summarizes how link failure recovery mayinfluence the network performance in two sets of valuesthe probabilities that the network is broken and the averagedistance of the network If network breaks the system maynot work If the average distance increases the systemrsquosperformance decreases because communication flows haveto take more hops to the destination In Figures 3(a) and3(b) the result shows that when there are more failed linksto be fixed (119877119886119905119894119900 119891119897119900119908 = 05) the system is in a higher

danger to be disconnected However a scale free networkperforms the worst and a grid network performs the bestThe reason is that if a hub agent is congested the networkis more likely to be disconnected On the other hand agentsin a grid network only connect to each other locally thenetwork is less likely to break down even when more andmore links or agents are congested Figures 3(a) and 3(b)also show that the average distance slowly increases in allnetwork topologies and before the network is broken downthe scale free network keeps the shortest distance and reservesthe small world effect best Figures 3(c) and 3(d) illustratethat when the average flow increases (119877119886119905119894119900 119897119894119899119896 = 002)


0005 0010 0015 0020 0025 0030 0035 0040 0045 00500

50

100

150

200

250

300

350

400

450

500

Ratio node

Num

ber o

f con

geste

d lin

ks


(a) Link congestion (vary percentage of failed node)


Ratio node0005 0010 0015 0020 0025 0030 0035 0040 0045 0050

0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des

(b) Node congestion (vary percentage of failed node)

Ratio flow

0

50

100

150

200

250

300

350

Num

ber o

f con

geste

d lin

ks

01 02 03 04 05 06 07 08 09



Ratio flow01 02 03 04 05 06 07 08 09

minus20

0

20

40

60

80

100

120

140

160

180


Num

ber o

f con

geste

d no

des


Figure 4 Congestions from node failure recovery by varying Ratio node and Ratio flow

a similar conclusion can be reached as in Figures 3(a) and3(b)

52 Robustness on Node Failure Recovery In this sectionwe investigate the network performances after the networkrecovery policy recovered node failures In Figures 2(b)and 4(b) we varied the failed agents in the network from05 to 5 and we set 119877119886119905119894119900 119891119897119900119908 = 05 We foundthat the congested links increased quickly while congestedagents slowly increased As we expected scale free networksmade heavy congested nodes Unlike Figures 2(a) and 4(a)although random and scale free networks have more numberof congested links when the failed nodes are sparse small

world and grid networks create about 40 more congestedlinks when failed nodes are more than 35 of all the agents

Figures 4(c) and 4(d) show the results that when weset the failed agents to be fixed as 2 (119877119886119905119894119900 119899119900119889119890 =

002) and varied the average flow of each link from 10to 90 of the 119865max both congested agents and congestedlinks are increasing in different complex network topologiesConsistent with the results of link failure recovery a randomnetwork performs the worst according to congested linkswhile a small world network works the best On the otherhand aswe expected a scale free network always createsmorecongested nodes while a grid network creates the least

Figure 5 shows when either the rate of failed nodesincreases or the average flow rate increases the probability



0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node

(a) Connectivity (vary percentage of failed node)


0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan

ce(b) Average distance (vary percentage of failed node)


00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow


Figure 5 Connectivity and average distance by varying Ratio node and Ratio flow

that the network is broken increases In Figures 5(a) and5(b) when there are more failed nodes to be recovered(119877119886119905119894119900 119891119897119900119908 = 05) the system is in a higher danger tobe disconnected however a scale free network performs theworst and a grid network performs the bestThe reason is thatif any hub agents are congested the network is much easierto be disconnected Figures 5(a) and 5(b) also show that theaverage distance slowly increases in all network topologiesand the scale free network still keeps the shortest averagedistance as we expected Figures 5(c) and 5(d) represent thatwhen the average flow increases (119877119886119905119894119900 119899119900119889119890 = 002) we canmake the similar conclusions as Figures 5(a) and 5(b) All the

experiments in this section are based on the setting of 119898 gt 0

(there are 15 backup links) but we could reach the sameconclusion when we set119898 = 0

53 Data Loss in Network Recovery As explained when themultiagent system comes to network failures althoughMPLShelps to reroute the data tomaintain the system performancenetwork congestions in the nodes or the links between themwill still bring communication loss In this subsection weinvestigate the percentages of data loss when the multiagentsystem is organized as different complex networks


01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure

Figure 6 Data loss from network recovery in different network topologies

In the first experiment we briefly use the basic settingof Section 51 that in the multiagent system there are 2links that are broken and the original average data volumein each link is 50 of 119865max When the system was organizedas four different complex network topologies we measuredthe percentages of the data loss in different scales The resultsare illustrated as in Figure 6(a) We can see that no matterthe system size is the grid network and small world networkmaintain good performances in link failure recoveries and thedata loss rates keep the lowest Consistent with our analysisin Section 51 the scale free network keeps a higher data lossrate and random network performs the worst in link failurerecovery where its data loss rate closes to 60

In the second experiment we briefly use the basic settingof Section 52 and there are 2 agents that are lost Whenthe system was organized as four different complex networktopologies Figure 6(b) briefly shows that no matter thesystem size is scale free network occupies the stability ofhub nodes and always performs the best and it is especiallythe case when the network scales up Consistent with ourconclusion in Section 52 small world network and randomnetwork bring out close performances while grid networkmade the worst data loss in node failure recovery In somecases the data loss rates are more than 70

6 Network Shifts on Recovery fromDifferent Topologies

In this section we verify if network recovery operationwouldlead to the changes of complex network topologies Similarexperiment settings are kept as Section 5 and both 119898 gt

0 (consists of 15 backup links) and 119898 = 0 are testedDuring the experiments we found that network recovery

usually does not lead to distinct shifts of network topologiesHowever when the number of congested links and nodesrapidly increases network connectivitymay be destroyedWeset119862max(119894) = 120582times119865max where120582 gt 1 is a constant so that agentsrsquocommunication volumes are fixed Our experiments areconducted by varying the parameters Ratio link Ratio nodeand Ratio flow but we always maintain that the network con-nectivity is not broken (very few results with disconnectednetworks are excluded) The experimentrsquos results are shownas degree distribution Each graph represents one type ofcomplex network and consists of three curves with threesettings the original network topology before any failures(Normal) the network topology after network recovery (119898 gt

0) and the network topology after network recovery withoutany backup links (119898 = 0)

61 Link Failure Recovery In this experiment we tested howthe different network topologies will be changed after linkfailure recovery Each network topology will be presented intwo different settings with different rate of link failure to berecovered (Ratio link) and different average flow (Ratio flow)which are very likely to break network connectivity

Figures 7(a) and 7(b) show how a random network topol-ogy shifts on two different settings Although the randomnetwork topology is kept and its distribution still follows aPoisson distribution the distribution clearly shifts left afterthe link failure recovered (it is more distinct when 119898 = 0)Therefore its average degree is decreased with link failurerecovery

Figures 7(c) and 7(d) show that the scale free networksignificantly shifted In both graphs the scale free networksare losing their power lawdistribution and aremore andmoreclose to a Poisson distribution as random networks In the


000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)

(a) Ratio link = 002 Ratio flow = 07

000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)

(b) Ratio flow = 05 Ratio link = 0035

p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)

(c) Ratio link = 002 Ratio flow = 06

p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

(d) Ratio flow = 05 Ratio link = 003

p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0

(e) Ratio link = 002 Ratio flow = 07

p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0

(f) Ratio flow = 05 Ratio link = 004

Figure 7 Network shifts from link failure recovery in different network topologies

settings of 119898 = 0 almost all the high degree agents are lostThe reason is that hub agents are easily congested whenmuchmore communication flow transmitted through hub agentsTherefore the small world effect is gradually disappeared

In Figures 7(e) and 7(f) the original small world networkbefore link recovery presents a generalized binomial distri-bution [29] However the network cannot keep this topologyin both graphs All the agents with higher degrees in a small


000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)

(a) Ratio node = 002 Ratio flow = 06

000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)

(b) Ratio flow = 05 Ratio node = 003

p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)

(c) Ratio node = 002 Ratio flow = 04

p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

(d) Ratio flow = 05 Ratio node = 001

p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0

(e) Ratio node = 002 Ratio flow = 06

p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0

(f) Ratio flow = 05 Ratio node = 003

Figure 8 Network shifts from node failure recovery in different network topologies

world network are more likely to be congested especially inthe settings of 119898 ge 0 Moreover when the average degreeof the networks decreases the average distance betweennodes rapidly increases and its degree distribution closes toa Poisson distribution after link failure recovery

62 Node Failure Recovery Similar to the experiments oflink failure recovery in Section 61 we vary two parametersof the node failure recoveries Ratio node and Ratio flowSimilar to link failure recovery schema networks are veryprone to be broken Figures 8(a) and 8(b) show the changes


on the random network Although the degree distributionsslightly change and the average degree decreases its Poissondistribution remains

Similar to the conclusion from Figures 7(c) and 7(d)Figures 8(c) and 8(d) show that the scale free networkscannot keep their topologies in both settings and aremore and more close to a random network However inFigure 8(c) the scale free network has a significant mutationwith the setting of 119898 gt 0 Its degree distribution hasbeen a combination of a Poisson distribution and a powerlaw distribution We found that the agent 119894 whose degreeis higher would present higher clustering coefficient 119862

119894

Otherwise low degree agents are loosely connected witheach other [30] Therefore the scale free network worksdistinctly between high clustering coefficient region and lowclustering coefficient region after node failures are recov-ered

Figures 8(e) and 8(f) illustrate that small world networkscannot keep their topologies in node failure recovery It issimilar to the phenomenon in Figures 7(e) and 7(f) In bothsettings degree distribution does not follow a generalizedbinomial distribution any more Therefore the average dis-tance between agents may be significantly increased

In conclusion the power law distribution in scale freenetwork and the binomial distribution in small network areunstable and can be easily destroyed by network congestionsOn the other hand a random network with Poissondistribution is stable Moreover network recovery operationby creating link or node congestions can significantlydecrease the average degree of the network In this sectionwe excluded the results from the grid network because itonly has local connections and the congestions have littleinfluences on its network topology

7 Conclusions and Future Work

Network recovery plays an important role in maintainingthe stability of the multiagent system in any applicationdomain However its impacts on different complex networkorganizations are still undiscovered In this paper we madeour initial efforts on studying those effects We have foundthat although the MPLS recovery mechanism can efficientlyrecover link or node failures by rerouting communicationflows via alternative paths it may bring node or link conges-tions on those alternative paths The congestions can signifi-cantly change the network topology and system performanceIn addition their effects are different on different complexnetworks By conducting extensive experiments we foundthat the small world effect and power law phenomenon ina scale free network are not stable in many cases Basedon our interesting discoveries we may be able to makelots of progresses in near future The first is to predict thesystem performance variances according to the changes ofthe network topology in multiagent coordination domainssuch as resource allocation information sharing and taskassignments Second we could optimize the recovery algo-rithm efficiency based on the utilizations of complex networkattributes

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

References

[1] C Wang B Wang G Zhang and Y Liang ldquoMethod offormation cooperative air defense decision based on multi-agent system cooperationrdquo inCommunications and InformationProcessing vol 289 of Communications in Computer and Infor-mation Science pp 529ndash538 2012

[2] I G Magarinoa and C Gutierrezb ldquoAgent-oriented modelingand development of a system for crisis managementrdquo ExpertSystems with Applications vol 40 no 16 pp 6580ndash6592 2013

[3] S Cranefield and S Ranathunga ldquoEmbedding agents in busi-ness applications using enterprise integration patternsrdquo inProceedings of the International Conference on AutonomousAgents and Multi-Agent Systems (AAMAS rsquo13) pp 1223ndash12242013

[4] M Musolesi S Hailes and C Mascolo ldquoAn Ad Hoc mobilitymodel founded on social network theoryrdquo in Proceedings of the7th ACM Symposium on Modeling Analysis and Simulation ofWireless and Mobile Systems (ACM MSWiM rsquo04) pp 20ndash24October 2004

[5] J Travers and S Milgram ldquoAn experimental study of the smallworld problemrdquo Sociometry vol 32 no 4 pp 425ndash443 1969

[6] A-L Barabasi and R Albert ldquoEmergence of scaling in randomnetworksrdquo Science vol 286 no 5439 pp 509ndash512 1999

[7] Y Xu M Lewis K Sycara and P Scerri ldquoInformation sharingin very large teamsrdquo in Proceedings of the 3rd InternationalJoint Conference onAutonomous Agent andMulti Agent Systems2004

[8] P Scerri and K Sycara ldquoSocial networks for effective teamsrdquo inCooperative Networks Control and Optimization Edward ElgarPublishing 2008

[9] M EGaston andMDesJardins ldquoAgent-organized networks fordynamic team formationrdquo inProceedings of the 4th InternationalConference on Autonomous Agents and Multi agent Systems(AAMAS rsquo05) pp 375ndash382 July 2005

[10] R Albert H Jeong and A-L Barabasi ldquoError and attacktolerance of complex networksrdquo Nature vol 406 no 6794 pp378ndash382 2000

[11] G A Pagani and M Aiello ldquoThe power grid as a complexnetwork a surveyrdquo Physica A Statistical Mechanics and ItsApplications vol 392 no 11 pp 2688ndash2700 2013

[12] Z Lin and K Carley ldquoDYCORP a computational frameworkfor examining organizational performance under dynamicconditionsrdquo Journal of Mathematical Sociology vol 20 no 2-3pp 193ndash217 1995

[13] Y C Jiang and J C Jiang ldquoUnderstanding social networks froma multi-agent coordination perspectiverdquo IEEE Transactions onParallel and Distributed Systems 2013

[14] M E Taylor M Jain Y n Jin M Yooko and M TambeldquoWhen should there be a ldquomerdquo in ldquoteamrdquo Distributed multi-agent optimization under uncertaintyrdquo in Proceedings of the 9thInternational Conference on Autonomous Agents andMultiagentSystems (AAMAS rsquo10) pp 109ndash116 2010

[15] T Liu X Li and X P Liu ldquoIntegration of small worldnetworks with multi-agent systems for simulating epidemicspatiotemporal transmissionrdquo Chinese Science Bulletin vol 55no 13 pp 1285ndash1293 2010


[16] L Gu X D Zhang and Q Zhou ldquoConsensus and syn-chronization problems on small-world networksrdquo Journal ofMathematical Physics vol 51 no 8 Article ID 082701 2011

[17] G DrsquoAngelo and S Ferretti ldquoSimulation of scale-free networksrdquoin Proceedings of the 2nd International Conference on SimulationTools and Techniques 2009

[18] X G Gong and J Xu ldquoResearch on delay characteristicsof information in scale-free networks based on multi-agentsimulationrdquo Procedia Computer Science vol 17 pp 989ndash10022013

[19] P Peschlow T Honecker and P Martini ldquoA flexible dynamicpartitioning algorithm for optimistic distributed simulationrdquoin Proceedings of the IEEE 21st International Workshop onPrinciples of Advanced and Distributed Simulation (PADS rsquo07)pp 219ndash228 San Diego Calif USA June 2007

[20] R Albert and A-L Barabasi ldquoStatistical mechanics of complexnetworksrdquo Review of Modern Physics vol 74 pp 47ndash97 2002

[21] P Erdos and A Renyi ldquoOn the evolution of random graphsrdquoPublications of the Mathematical Institute of the HungarianAcademy of Sciences vol 5 pp 17ndash61 1959

[22] D J Watts and S H Strogatz ldquoCollective dynamics of small-world networksrdquo Nature vol 393 no 6684 pp 440ndash442 1998

[23] L Bononi M Bracuto G DrsquoAngelo and L Donatiello ldquoExplor-ing the effects of hyper-threading on parallel simulationrdquoin Proceedings of the 10th IEEE International Symposium onDistributed Simulation and Real-Time Applications (DS-RT rsquo06)pp 257ndash260 Torremolinos Spain October 2006

[24] A Broder R Kumar F Maghoul et al ldquoGraph structure in thewebrdquo Computer Networks vol 33 no 1 pp 309ndash320 2000

[25] R K Merton ldquoThe Matthew effect in sciencerdquo Science vol 159no 3810 pp 56ndash63 1968

[26] R Cohen S Havlin and D Ben-Avraham ldquoStructural prop-erties of scale-free networksrdquo in Handbook of Graphs andNetworks 2003

[27] J Leskovec J Kleinberg and C Faloutsos ldquoGraphs over timedensification laws shrinking diameters and possible explana-tionsrdquo in Proceedings of the 11th ACM SIGKDD InternationalConference on Knowledge Discovery andDataMining (KDD 05)pp 177ndash187 August 2005

[28] V Sharma and F Hellstrand ldquoFramework for multi-protocollabel switching (MPLS)-based recoveryrdquo RFC 3469 2003

[29] K Mahdi H Farahat and M Safar ldquoTemporal evolution ofsocial networks in Paltalkrdquo in Proceedings of the 10th Interna-tional Conference on Information Integration and Web-BasedApplications and Services (iiWAS rsquo08) pp 98ndash103 November2008

[30] X Yao C S Zhang J W Chen and Y D Li ldquoOn theformation of degree and cluster-degree correlations in scale-freenetworksrdquo Physica A Statistical Mechanics and its Applicationsvol 353 no 1ndash4 pp 661ndash673 2005

International Journal of

AerospaceEngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

RoboticsJournal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Active and Passive Electronic Components

Control Scienceand Engineering

Journal of



RotatingMachinery


Hindawi Publishing Corporation httpwwwhindawicom

Journal ofEngineeringVolume 2014

Submit your manuscripts athttpwwwhindawicom

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics


Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

SensorsJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation



DistributedSensor Networks



hoc network Although the popular restoration of reroutingmechanism for exampleMPLS recovery algorithm has beenproven to be feasible on network recoveries the potentialshifts on the multiagent network topology may significantlychange the system performance in an unpredictable way

In our simulation we simulated a series of large scalemultiagent systems with different complex network topolo-gies A popular restoration of rerouting mechanism-MPLSalgorithm is implemented to recover link and node failuresThe network is evaluated by its diameter average distanceand cluster [10] To simulate the physical network in thoseexperiments the communication capabilities of each link ornode are limitedThe experiment results are presented in twomajor sections system robustness and influence of networktopologies

In the first section of the experiments data that flowsthrough the failure agents or links has to be rerouted andmay cause congestions on existing links or agents Basedon our discussion about the efficiency of network recoverythe number of newly congested links or nodes (agents) thathurt system performance is investigated More importantlythe network is in danger of being disconnected from thosecongestions By comparing the influence on different networktopologies while there are some differences in different casesour major discoveries are that a random network is morelikely to be congested if more links are broken while asmall world network is the least likely On the other handa scale free network appears to be more vulnerable to nodecongestions while a grid network is the most robust

In the second section we test how the network topologymay be shifted from the network recovery We have foundthat most of network topologies are immune to the networkrecovery when the congestions are not so serious Howeverwhen the number of congested links and nodes is highlyincreased the network topologies may have been changedIt is especially the case when the network is organized as ascale free network or a small world network However sincescale free and small world networks are the most importantattributes in a large multiagent system from our previousresearch experience their changesmay significantly affect thesystem performances

2 Related Work

The structure and behavior of complex networks have beenattractive in various studies [11] Inspired by the discoverieson how the rich network structure facilitates effective orga-nizational behavior [12] Gaston and DesJardins illustratedthe importance of network topologies in multiagents net-works [9] Y C Jiang and J C Jiang analyzed the complexnetwork in actor-oriented and actor-structure views to findthe relationship between complex networks and multiagentsystems [13] Taylor et al examined joint actions in themultiagent optimization problem and the results are sur-prising because a high number of connections in a complexnetworked multiagent team can hurt system performanceeven communication and computation costs are ignored [14]Liu et al proposed an integrated model based on small

Figure 1 Different complex network topologies random networkgrid network small world network and scale free network

world network and multiagent system to simulate epidemicspatiotemporal transmissionThey found that the small worldeffect brings better performance than the traditional modeland his discovery has been applied in real geographicalmultiagent applications [15]

On the other hand a set of researches show that net-work operations could influence the performance of thelarge scale multiagent systems as well For example whenfacing the same network attack the network vulnerabili-ties are different if the multiagent systems are organizedas various types of complex networks [16] DrsquoAngelo andFerretti explained that gossip protocols would change thecommunication of the complex network and have someimpacts on routing efficiency [17] Gong andXu analyzed andtested that different parameters on a scale free network canmake significant different efficiencies of information delayin multiagent systems [18] Peschlow et al [19] described aflexible dynamic partitioning algorithm that rapidly recoversinformation routing and optimizes the performance withdifferent complex network topologies

3 Modeling the Complex Networks

The network topology of a multiagent system is defined asan undirected graph 119866 = (119881 119864) where 119881 = 1 2 119873119873 = |119881| defines the agent set 119864 denotes the set of linksbetween agents that is link (119894 119895) isin 119864 and 119894 and 119895 areneighbors 119899 119881 rarr 119864 defines the set of all neighbors of anagent That is 119899(119894) = 119887 119895 119897 is the neighbor of agent 119894 119866could be organized as different network topologies based onthe different properties of complex networks In this paperwe are mainly interested in four of them shown in Figure 1random network grid network small world network andscale free network Preliminary studies [20] have found thateach topology encodes different fundamental properties thatis network diameter average distance between nodes clusterand degree distributions

(i) Degree the degree of agent 119894 is 119889(119894) = |119899(119894)|

(ii) Average degree 119889 = (1119873)sum119894isin119881

|119899(119894)| is the averagenumber of neighbors of all agents for any complexnetwork 119889 ≪ |V|

(iii) Degree distribution 119901(119896) = 119875119903[119889 = 119896] defined as afraction of agents (the number of such agents is 119889)with the degree 119896

(iv) Distance distance(119894 119895) is defined as the least numberof hops to communicate between the agents 119894 and 119895Specifically distance(119894 119895) = 1 if (119894 119895) isin 119864



119897 =1

119873 (119873 + 1)sum

forall119894119895isin119881

distance (119894 119895) (1)






119862119894=

2 |119890 (119894)|

|119899 (119894)| lowast (|119899 (119894)| minus 1) (2)


119873

119894=1119862119894


minus119903






1015840 1198641015840) where 119881



(3) 119878119890119899119889119863119886119905119886(119908119891 (119894 119895)

|119904 (119894)|)

(4) end for

Algorithm 1 119871119894119899119896 119877119890119888119900V119890119903119910((119894 119895) 119891(119894 119895))


Algorithm 2 119860119892119890119899119905 119877119890119888119900V119890119903119910(119894)




1015840)









Algorithm 3 119862119900119899119892119890119904119905119894119900119899(119894 119895)












Ratio link

300

250

200

150

100

50

00005 0010 0015 0020 0025 0030 0035 0040 0045 0050

Num

ber o

f con

geste

d lin

ks



Ratio link0005 0010 0015 0020 0025 0030 0035 0040 0045 0050


0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

Num

ber o

f con

geste

d lin

ks


01 02 03 04 05 06 07 08 09

0

50

100

150

200

250


Ratio flow01 02 03 04 05 06 07 08 09


0

20

40

60

80

100

120

140

Num

ber o

f con

geste

d no

des











Ratio link0005 0010 0015 0020 0025 0030 0035

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

00



Ratio link00050000 0010 0015 0020 0025 0030 0035

0

5

10

70

80

90

100

Aver

age d

istan

ce


Ratio flow

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05 06 07




Ratio flow

0

5

10

70

80

90

100Av

erag

e dist

ance

00 01 02 03 04 05 06 07







0005 0010 0015 0020 0025 0030 0035 0040 0045 00500

50

100

150

200

250

300

350

400

450

500

Ratio node

Num

ber o

f con

geste

d lin

ks




Ratio node0005 0010 0015 0020 0025 0030 0035 0040 0045 0050

0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

0

50

100

150

200

250

300

350

Num

ber o

f con

geste

d lin

ks

01 02 03 04 05 06 07 08 09



Ratio flow01 02 03 04 05 06 07 08 09

minus20

0

20

40

60

80

100

120

140

160

180


Num

ber o

f con

geste

d no

des











0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node



0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan



00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow








01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











119897 =1

119873 (119873 + 1)sum

forall119894119895isin119881

distance (119894 119895) (1)






119862119894=

2 |119890 (119894)|

|119899 (119894)| lowast (|119899 (119894)| minus 1) (2)


119873

119894=1119862119894


minus119903






1015840 1198641015840) where 119881



(3) 119878119890119899119889119863119886119905119886(119908119891 (119894 119895)

|119904 (119894)|)

(4) end for

Algorithm 1 119871119894119899119896 119877119890119888119900V119890119903119910((119894 119895) 119891(119894 119895))


Algorithm 2 119860119892119890119899119905 119877119890119888119900V119890119903119910(119894)




1015840)









Algorithm 3 119862119900119899119892119890119904119905119894119900119899(119894 119895)












Ratio link

300

250

200

150

100

50

00005 0010 0015 0020 0025 0030 0035 0040 0045 0050

Num

ber o

f con

geste

d lin

ks



Ratio link0005 0010 0015 0020 0025 0030 0035 0040 0045 0050


0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

Num

ber o

f con

geste

d lin

ks


01 02 03 04 05 06 07 08 09

0

50

100

150

200

250


Ratio flow01 02 03 04 05 06 07 08 09


0

20

40

60

80

100

120

140

Num

ber o

f con

geste

d no

des











Ratio link0005 0010 0015 0020 0025 0030 0035

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

00



Ratio link00050000 0010 0015 0020 0025 0030 0035

0

5

10

70

80

90

100

Aver

age d

istan

ce


Ratio flow

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05 06 07




Ratio flow

0

5

10

70

80

90

100Av

erag

e dist

ance

00 01 02 03 04 05 06 07







0005 0010 0015 0020 0025 0030 0035 0040 0045 00500

50

100

150

200

250

300

350

400

450

500

Ratio node

Num

ber o

f con

geste

d lin

ks




Ratio node0005 0010 0015 0020 0025 0030 0035 0040 0045 0050

0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

0

50

100

150

200

250

300

350

Num

ber o

f con

geste

d lin

ks

01 02 03 04 05 06 07 08 09



Ratio flow01 02 03 04 05 06 07 08 09

minus20

0

20

40

60

80

100

120

140

160

180


Num

ber o

f con

geste

d no

des











0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node



0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan



00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow








01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Algorithm 3 119862119900119899119892119890119904119905119894119900119899(119894 119895)












Ratio link

300

250

200

150

100

50

00005 0010 0015 0020 0025 0030 0035 0040 0045 0050

Num

ber o

f con

geste

d lin

ks



Ratio link0005 0010 0015 0020 0025 0030 0035 0040 0045 0050


0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

Num

ber o

f con

geste

d lin

ks


01 02 03 04 05 06 07 08 09

0

50

100

150

200

250


Ratio flow01 02 03 04 05 06 07 08 09


0

20

40

60

80

100

120

140

Num

ber o

f con

geste

d no

des











Ratio link0005 0010 0015 0020 0025 0030 0035

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

00



Ratio link00050000 0010 0015 0020 0025 0030 0035

0

5

10

70

80

90

100

Aver

age d

istan

ce


Ratio flow

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05 06 07




Ratio flow

0

5

10

70

80

90

100Av

erag

e dist

ance

00 01 02 03 04 05 06 07







0005 0010 0015 0020 0025 0030 0035 0040 0045 00500

50

100

150

200

250

300

350

400

450

500

Ratio node

Num

ber o

f con

geste

d lin

ks




Ratio node0005 0010 0015 0020 0025 0030 0035 0040 0045 0050

0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

0

50

100

150

200

250

300

350

Num

ber o

f con

geste

d lin

ks

01 02 03 04 05 06 07 08 09



Ratio flow01 02 03 04 05 06 07 08 09

minus20

0

20

40

60

80

100

120

140

160

180


Num

ber o

f con

geste

d no

des











0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node



0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan



00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow








01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










Ratio link

300

250

200

150

100

50

00005 0010 0015 0020 0025 0030 0035 0040 0045 0050

Num

ber o

f con

geste

d lin

ks



Ratio link0005 0010 0015 0020 0025 0030 0035 0040 0045 0050


0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

Num

ber o

f con

geste

d lin

ks


01 02 03 04 05 06 07 08 09

0

50

100

150

200

250


Ratio flow01 02 03 04 05 06 07 08 09


0

20

40

60

80

100

120

140

Num

ber o

f con

geste

d no

des











Ratio link0005 0010 0015 0020 0025 0030 0035

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

00



Ratio link00050000 0010 0015 0020 0025 0030 0035

0

5

10

70

80

90

100

Aver

age d

istan

ce


Ratio flow

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05 06 07




Ratio flow

0

5

10

70

80

90

100Av

erag

e dist

ance

00 01 02 03 04 05 06 07







0005 0010 0015 0020 0025 0030 0035 0040 0045 00500

50

100

150

200

250

300

350

400

450

500

Ratio node

Num

ber o

f con

geste

d lin

ks




Ratio node0005 0010 0015 0020 0025 0030 0035 0040 0045 0050

0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

0

50

100

150

200

250

300

350

Num

ber o

f con

geste

d lin

ks

01 02 03 04 05 06 07 08 09



Ratio flow01 02 03 04 05 06 07 08 09

minus20

0

20

40

60

80

100

120

140

160

180


Num

ber o

f con

geste

d no

des











0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node



0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan



00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow








01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











Ratio link0005 0010 0015 0020 0025 0030 0035

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

00



Ratio link00050000 0010 0015 0020 0025 0030 0035

0

5

10

70

80

90

100

Aver

age d

istan

ce


Ratio flow

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05 06 07




Ratio flow

0

5

10

70

80

90

100Av

erag

e dist

ance

00 01 02 03 04 05 06 07







0005 0010 0015 0020 0025 0030 0035 0040 0045 00500

50

100

150

200

250

300

350

400

450

500

Ratio node

Num

ber o

f con

geste

d lin

ks




Ratio node0005 0010 0015 0020 0025 0030 0035 0040 0045 0050

0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

0

50

100

150

200

250

300

350

Num

ber o

f con

geste

d lin

ks

01 02 03 04 05 06 07 08 09



Ratio flow01 02 03 04 05 06 07 08 09

minus20

0

20

40

60

80

100

120

140

160

180


Num

ber o

f con

geste

d no

des











0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node



0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan



00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow








01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










0005 0010 0015 0020 0025 0030 0035 0040 0045 00500

50

100

150

200

250

300

350

400

450

500

Ratio node

Num

ber o

f con

geste

d lin

ks




Ratio node0005 0010 0015 0020 0025 0030 0035 0040 0045 0050

0

10

20

30

40

50

60

70

80

Num

ber o

f con

geste

d no

des


Ratio flow

0

50

100

150

200

250

300

350

Num

ber o

f con

geste

d lin

ks

01 02 03 04 05 06 07 08 09



Ratio flow01 02 03 04 05 06 07 08 09

minus20

0

20

40

60

80

100

120

140

160

180


Num

ber o

f con

geste

d no

des











0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node



0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan



00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow








01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation











0005 0010 0015 0020 0025

00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

Ratio node



0005 0010 0015 0020 0025

Ratio node00000

5

10

70

80

90

100

Aver

age d

istan



00

02

04

06

08

10

Prob

abili

ty o

f disc

onne

cted

net

wor

ks

01 02 03 04 05

Ratio flow



0

5

10

70

80

90

100

Aver

age d

istan

ce

00 01 02 03 04 05

Ratio flow








01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










01

02

03

04

05

06

07

08

09

Dat

a los

s (

)

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000


(a) Link failure

System scale500 1000 1500 2000 2500 3000 3500 4000 4500 5000

02

03

04

05

06

07

08

09

10


Dat

a los

s (

)(b) Node failure













000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 12

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0






000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation










000

005

010

015

020

025

2 4 6 8 10 12 14 16

p(k)

Node degree (k)


000

005

010

015

020

025

p(k)

1 2 3 4 5 6 7 8 9 10 11 1312

Node degree (k)


p(k)

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30000

005

010

015

020

025

030

Node degree (k)


p(k)

000

005

010

015

020

025

030

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0


p(k)

000

005

010

015

020

025

030

035

040

2 4 6 8 10 12 14 16 18 20 22 24 26 28 30

Node degree (k)

Normalm gt 0m = 0








119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation












119894








References


































RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation



























RoboticsJournal of





Journal of



RotatingMachinery





VLSI Design



Shock and Vibration







Journal of



Volume 2014


SensorsJournal of





Propagation









Research Article Discovering the Influences of Complex...

Documents

Transcript of Research Article Discovering the Influences of Complex...