ĐIỆN tử VIỄN THÔNG pp17 an optimized scheme to organize softswitch based NGN khotailieu

First, a non-linear pro-gramming scheme that aims to minimize the cost of organizing networks under the constraint of required reliability and delay is analyzed, and we present a heuris

An Optimized Scheme to Organize Softswitch-based Next Generation Network1

Yin Zeming

State Key Laboratory of Switching Technology and Telecommunication Networks, Beijing University of Posts and Telecommunications, Box 187, Beijing 100876, China

yinzeming2002@163.com

Liu Yuzhang

liuyz@mail.dascom.com.cn

Yang Fangchun

fcyang@bupt.edu.cn

Abstract: This paper outlines the network organizing in softswitch-based NGN First, a non-linear

pro-gramming scheme that aims to minimize the cost of organizing networks under the constraint of required reliability and delay is analyzed, and we present a heuristic algorithm to solve it The reliability of the net-work is guaranteed with both the reliability of the links and that of the softswitch nodes considered, and

‘state-sharing’ is adopted to assure the node’s reliability A design scheme of softswitch node supporting

‘state-sharing’ is presented Finally, the evaluation for the heuristic algorithm is presented

Keywords: NGN, softswitch, non-linear programming, heuristic algorithm

1 INTRODUCTION

Now it is a trend that the various networks running independently will converge to Next Generation Network (NGN), and softswitch-based scenario is one of the effectual schemes Accordingly, organizing scheme and call routing become important issues to be researched

However, few researches have focused on organizing technology and call routing of softswitch-based NGN so far NGN is all-IP network, and is operates in a manageable and operable mode rather than the best-effort mode in current Internet At the same time, while Public Switched Telephone Network (PSTN) has accumulated rich solutions in organizing networks, it is based on circuit-switch architecture whose ser-vice traffic doesn’t behave like that in the data networks Up to now there is no determined scheme for softswitch-based NGN’s organizing in the related standards

This remainder of the paper is organized as follows In Section 2, we describe NGN’s objectives in organizing network, and analyze them from the economic view and the performance view We present a model based on the foundation that the average delay in data networks is got through Kleinrock Approxi-mation [1] and Jackson theorem A heuristic algorithm and its related heuristic strategies are presented to

1

The work is supported by National Nature Science Fund (90104024), National Science Fund for Distinguished Young Scholars (60125101)

V.B IVERSEN and KUO G.S.(Editors)

Beijing University of Posts and Telecommunications Press

solve the model Section 3 describes a state-sharing scheme with which to increase the reliability of softswitch node Finally, the heuristic algorithm is evaluated in Section 4

There are a few schemes that can be adopted to organize NGN, such as hierarchical, non-hierarchical

or mixed In the hierarchical way, softswitch works like switches in PSTN In the non-hierarchical way, softswitch works in the same way as routers in IP networks In [2], a mixed scheme is presented In the scheme, the non-hierarchical architecture of softswitch is revised a little with Routing Agent added There are other schemes to organize NGN, and here we only discuss the non-hierarchical scheme Our objective

in organizing the network includes:

Keeping the average delay of each packet or each message under a certain level

Raising the reliability of a single softswitch node

Minimizing the cost under the condition that the two requirements above are met

It is assumed that the location of the softswitch node and the traffic are known, and our aim is to or-ganize the network meeting the requirement in traffic and performance with least cost It is a difficult and synthesized problem Considering the complexity, we launch on minimizing the cost by programming the capacity of the links between softswitches The problem of organizing network is boiled down to program-ming the capacity of each link (which is described as (i, j)),

T F C

F

t

s

C

p

ij

j

ij

ij

≤

−

∑

∑

)

(

)

(

1

.

.

min

γ

(1)

where Cijis the capacity of the link ( j i, ), pijis the cost of each unit capacity, and the constraint

condi-tion is that the average delay is no larger than T Fij is the flow of the link ( j i , ) holding the same

) (

1

ij F C

F

forcast, and γ is the total arriving rate into the network In[2], it is proved that Kleinrock forcast is an ef-fective framework to approximate the average delay of packet in data networks and the forcast fits well with the Jackson theorem Here we assume that a routing scheme is assigned originally and so the flow Fij

is known As far as the expression (1) is concerned, it is clear that the minimum can be reached when the restriction is an equation Here we introduce a Lagrange multiplicator β and construct the Lagrange

) (

) (

ij ij

ij

F C

F C

p L

γ

β

According to the knowledge of Lagrange product, we assume

that the derivative is equal to zero

0 )

−

−

=

∂

∂

ij ij

ij ij

F p

C

L

γ

β

and we can get

(2)

ij ij ij ij p F F C γ β + = put it into the constraint (1), and we can get ∑ ∑ = − = ) ( ) ( 1 j ij ij j ij ij ij p F F C F T βγ γ and so (3)

1 2 ) ( ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ = ∑ j ij ijF p T γ β put it into the equation (2), and we can get ∑ + = ) ( 1 n m nm nm ij ij ij ij F p p F T F C γ γ that is, (4)

1 1 ( , ) ⎟⎟ ⎟ ⎟ ⎠ ⎞ ⎜⎜ ⎜ ⎜ ⎝ ⎛ + = ∑ ij ij n m mn mn ij ij F p F p T F C γ Finally, put it into the cost function, and we can get the optimized cost (5)

)

(

) , )

=

j

ij ij j

ij ij

T F p

C

γ

And now it is time to consider how to optimize the network according to the concrete flow Fij The

prob-lem can be solved by optimizing expression (5) to get Fij, however, (5) has so many local minimums that

it is difficult to get the global minimum A more boring problem is that Fij and its corresponding Cij

in-cline to zero in the local minimum, which may make the traffic focus on a small quantity of links, and thus decreases the reliability of the network

According to the reasoning above, it is difficult to optimize Fij and Cij at the same time And even

if the problem can be solved, the traffic will focus on some links with large capacity, which will lead to the condition that the requirement on the reliability of network can’t be met At the same time, the cost of the capacity is not linear to the capacity strictly speaking Thus, the only feasible method to solve the problem meeting the constraint is to adopt some heuristic algorithms Here we introduce a prototype of the iterative

and heuristic algorithms Generally speaking, the algorithms begin with an existing topology, and iterates in changing the topology by modifying the capacity of one or more links At the beginning of iteration, a topology exists and after the iteration a new topology that can meet the requirement of delay and reliability but pay a lower cost is found We take the following assumptions:

1 A predicted traffic demand exists;

2 The demand and an routing model have determined a feasible topology In the topology, each link ca-pacity is expressed as Cij, and the cost function is expressed as ∑

) (

) (

j ij

ij F

D , and the Dij can be

described as the following equation:

(6)

) ( ij ij ij ij ij ij ij d F F C F F D + − = 3 Then F can be determined through minimizing the average packet delay ij (7)

1

) (

⎠

⎞

⎜

⎜

⎝

⎛

+

−

=

j

ij ij ij ij

ij

F d F C

F D

γ

4 A constraint on delay must be met Generally speaking, the delay got through (7) is required to be no larger than a certain threshold value;

5 A constraint on reliability must be met For example, a k-connection network is required, that is, even if there are k − 1 nodes that are disabled, the other nodes are all reachable;

6 For a concrete evaluated network, there is a standard for cost;

We are looking for a topology that meet the requirement of 3 and 4 mentioned above and pay a lowest cost in 5 for it Here we present a heuristic algorithm working by iteration At the beginning of each itera-tion, there is a currently best topology and a topology to be tried The former meets the requirement on delay and reliability, and pays the least cost as far; the latter is the one to be evaluated in this iteration Here

we assume that the original topology has been chosen by some special method The steps for iteration is the following:

Step 1: Assigning the flow; calculate the flow Fij on the link ( j i , ) with some special routing algorithm;

Step 2: Inspecting the delay; estimate the average delayDunder the condition of current topology with equation (7), and if D ≤ T(Tis a threshold) then proceed with step 3, else proceed with step 5;

Step 3: Inspecting the reliability; test whether the trial topology meets the requirement on reliability, and if

the requirement is met then proceed with step 4, else proceed with step 5;

Step 4: Inspecting the amelioration of cost; replace the currently best topology with the trial topology if the

cost of the trial topology is less than that of the currently best one;

Step 5: Generating a new trial topology; generate a new trial topology generated by changing the capacity

of one or more links with some heuristic methods, and then proceed with step 1 for another iteration;

Note that only when the trial topology in Step 4 should meet the requirement in Step 2 and Step 3, and the cost should be ameliorated, the trial topology can be accepted as the currently best topology The algo-rithm ends when no new trial topology will generate or when the cost can be ameliorated obviously In fact, the algorithm can’t guarantee that the final result is best; a method that can improve the result is that a new originating topology can be tried to initiate the algorithm

Here we present a heuristic rule to generate new heuristic topology in step 5 An approach is to decrease the capacity of a link whoseF ij C ij is very low, or even withdraw the link Another possible approach is to

in-crease the capacity of the links whose F ij C ij is so large that the requirement on delay can’t be met Thus,

some links may be added or be deleted Here we present a method called saturation-cut, which aims to de-termine the partition zone between two sets,N1andN2 The links of the partition zone is utilized high

effi-ciently, and adding a link between N1 and N2 to lighten the high utility rate The working flow of the method is the following:

Step 1: List all the non-directional links, and arrange them according to the value of

) ,

max( Fij Cij Fji Cji in a descending order;

Step 2: Look for a link k to satisfy the following two conditions: a) if all the links above k are deleted, the network is also connected; b) if all the links above k and k are deleted, the network is non-connected and divided into two parts,N1andN2;

Step 3: Delete the least utilized links, and add a new link connecting a node inN1and a node inN2 Fig.1 shows an example of the Saturation Cut strategy The number beside each arc of the graph is the utility rate of the link According to Fig.1, high utility rate links are removed temporarily until the network

is divided into two parts, N1 and N2 When a

low utility rate link is deleted, a new link

connect-ing a node in N1and a node inN2 is added

There are also some other strategies to

gener-ate topologies for the step 4 of the heuristic

algo-rithm For example, when selecting the link to be

deleted, we can take ∑

) ( j

ij

ijC

3 THE RELIABILITY OF SOFTSWITCH

NODE

The reliability of the links has been ensured

4

7

5

8

3

2

1

11

10

0.8

0.7

0.75

0.6 0.5

0.3

0.5

0.4 0.75 0.3

0.5 0.2

0.6 0.3

0.4

1

Saturation Cut

Fig 1 An example of the Saturation Cut

through the algorithm above Here we are taking measures to raise the reliability of softswitch nodes A fault-tolerant system is required to achieve the goal [3][4] In our scheme, state-sharing technology is adopted to provide reliable services in softswitch nodes In [5][6], state-sharing technology is introduced and applied in a SIP-based system Here in designing our scheme of deploying softswitch nodes, we use the state-sharing model for reference The scheme is shown as the following figure

In the scheme shown as Fig.2, a text file is used as a medium for transporting SIP state update mes-sages The scheme of a XXX system that implements the state-sharing mechanism consists of three com-ponents on each host

STC (SUM-to-txt converter)

FTS (File transfer script)

TSC (Txt-to-SUM converter)

The XXX server may be a SIP proxy

server, a H.323 server or any other function

deployed in softswitch node The STC and

FTS are running in the server that

gener-ates/updates a state (source, i.e., host1),

while TSC is running in the server

(destina-tion, i.e., host2) that receives the text file

containing the state update message (SUM)

The SUM is generated in the message handler (MH)

and added to the input queue of the state manager

(SM) The STC takes a SUM from the MH as input and generates a text file as output The failure-detection and the fail-over management mechanisms of state-sharing technology are described in [6]

According to the result of the experiment in [7], when a one-proxy scenario’s reliability is 98.95%, the reliability of the two-proxy scenario adopting the state-sharing technology is 99.98%

With the node’s reliability provided with state-sharing, we assume that the reliability of the softswitch nodes is adequately ensured And in the concrete example to be shown, we only consider the reliability of the links And we assume that the links’ reliability can be ensured if each node has no less than two nodes

as its neighbors The T in expression (1) is assumed to be 2 seconds The initial cost shown in Fig.1 is

3542 The price matrixPand the flow matrixFis shown as the following:

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎦

⎤

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎣

⎡

=

0 4 4 5 4 8 8 9 11 11 12

4 0 7 6 4 10 9 8 13 12 13

4 7 0 4 6 4 6 9 8 9 10

5 6 4 0 3 5 4 5 8 8 9

4 4 6 3 0 8 5 4 10 8 10

8 10 4 5 8 0 3 7 4 6 7

8 9 6 4 5 3 0 4 4 3 6

9 8 9 5 4 7 4 0 7 3 8

11 13 8 8 10 4 4 7 0 4 3

11 12 9 8 8 6 3 3 4 0 4

12 13 10 9 10 7 6 8 3 4 0

P

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎥

⎦

⎤

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎢

⎣

⎡

=

0 15 15 0 0 0 0 0 0 0 0

18 0 0 0 25 0 0 0 0 0 0

12 0 0 14 0 0 0 0 0 0 0

0 0 16 0 20 0 24 0 0 0 0

0 24 0 19 0 0 0 22 0 0 0

0 0 0 0 0 0 14 0 18 0 0

0 0 0 24 0 14 0 17 20 16 0

0 0 0 0 20 0 25 0 0 20 0

0 0 0 0 0 21 17 0 0 0 22

0 0 0 0 0 0 18 16 0 0 19

0 0 0 0 0 0 0 0 28 25 0

F

LAN

Host 1

STC

FTS

TSC

XXX Server

Text file SUM

Host 2

STC

FTS

TSC

XXX Server

Text file SUM

Fig 2 The state-sharing scheme

The routing method that we adopt in the algorithm is the N algorithm And the service traffic matrix is

omitted here The results according to the strategies are the following:

Fig 3 The result with heuristic strategy one

The cost in Fig.3 is 3156, and the cost in Fig.4 is 3095 The comparison between Fig.3 and Fig.4 shows that, if max( Fij Cij, Fji Cji) is to be considered in the strategy, the utility rate in the links is

very close, and if the cost is to be considered when deleting the low utility link, the flow will concentrate

on the high capacity links which locates in the center of the network

The heuristic algorithm is not the best optimized one, but it is scalable to various strategies with which the behavior of the network can be adjusted That is just what is needed in operating NGN

REFERENCES:

[1] M Gerla, L Kleinrock, “On the Topological Design of Distributed Computer Networks”, IEEE Trans-action on Communications, Volume: 25, Issue: 1, Pages: 48 – 60, Jan 1977

[2] Z.G Wang, S Su, and J.L Chen, “A Research on the Call Routing of Softswitch”, Proceedings of ICCT2003, vol.2, Pages: 1594 – 1597, 2003

[3] A Helal, A Heddaya, and B Bhargava, Replication Techniques in Distributed Systems, Kluwer Aca-demic Publishers, 1996

[4] G Coulouris, J Dollimore, and T Kindberg, Distributed Systems: Concepts and Design, 3rd Edition Addison Wesley, 2001

[5] M Bozinovski, L Gavrilovska, and R Prasad, “Report on State-sharing Concepts”, Internal Siemens -CPK project report, February 2002

[6] M Bozinovski, L Gavrilovska, and R Prasad, “Performance evaluation of a SIP-based state- sharing mechanism,” Proceedings of Vehicular Technology Conference, vol.4, Pages: 2041-2045, 2002

[7] M Bozinovski, L Gavrilovska, and R Prasad, “A State-sharing Mechanism for Providing Reliable SIP Sessions, Telecommunications in Modern Satellite, Cable and Broadcasting Service,” vol.1, Pages: 384-

388, 2003

4

7

5

8

3

2

1

11

10 0.6

0.75

0.75

0.6 0.65

0.6 0.6

0.7

0.75 0.7

0.6

0.6 0.6

0.7

0.6

Fig 4 The result with heuristic strategy two

4

7

5

8

3

2

1

11

10 0.8

0.8

0.7

0.5 0.7

0.7 0.5

0.8 0.7

0.6

0.6 0.6

0.5 0.5