Available online at www.sciencedirect.

com

Procedia Engineering 15 (2011) 3341 – 3345

Advanced in Control Engineering and Information Science

Optimization of Cloud Database Route Scheduling Based on

Combination of Genetic Algorithm and Ant Colony Algorithm
ZHANG Yan-huaa, Feng Leia, Yang Zhia, a*
Institute of Information Engineering, Shenyang University of Chemical Technology, Shenyang China

Abstract

For the cloud database route scheduling problem, this paper designed a cloud database route scheduling algorithm
according to the dynamic combination of the genetic algorithm and ant colony algorithm. The initial solution got by the
Genetic Algorithm was transformed into the pheromone initial value, which was needed by ant colony algorithm, then
the optimal solution by the ant colony algorithm was obtained .Genetic control function was set up to control the
opportunity of two algorithm's fusion .This paper proposed a reasonable algorithm ,which could find the required
database rapidly and effectively, reduce the dynamical load of cloud database routing, and improve the efficiency of
cloud computing.
© 2011 Published by Elsevier Ltd. Selection and/or peer-review under responsibility of [CEIS 2011]
Key words: Genetic algorithm˗Ant colony algorithm˗Cloud database˗Dynamic routing scheduling;

1. Introduction

Cloud computing is a computer network model, which is combined by the traditional computer
technology˄such as Grid Computing, Distributed Computing, Parallel Computing, Utility Computing,
Network Storage Technology, Virtualization, Load Balance and so on˅and the network technology. It
integrates multiple low cost computers into a powerful computing capability system through the network,
and distribute these powerful computing capabilities among those terminal users by business models, such
as SaaSǃPaaSǃIaaSǃMSP. Cloud database is the data storage and retrieval resource of the cloud computing.
Thus, how to quickly and reasonably find out the best suited to applications's database is always the
different point.

*
Corresponding author. Tel.: 13998263712; fax:024-83995221
E-mail address: [email protected]

1877-7058 © 2011 Published by Elsevier Ltd.

doi:10.1016/j.proeng.2011.08.626
3342 ZHANG Yan-hua et al. / Procedia Engineering 15 (2011) 3341 – 3345

2 Cloud database

Cloud database system is composed of several sites, which are also called notes. Notes are linked in
the communication network, and each one has its own database, central processing unit (CPU), terminal
and local database management system (LDMS). There for, cloud database system can be regarded as a
union of a series of centralized database systems, which are unified logically, but distributional physically.
Cloud database consists of an immense amount of data systems. A part of them are stored on the cloud,
while some others are constituted by some small and medium-sized service providers's servers. This part of
servers will continuously join or exit from the cloud with the operational state of these companies. Sudden
failures may occurs in nodes and links, and new node or link may also connect to the cloud database at all
times. The distributivity, dynamics, global balance and scalability of the cloud database make the cloud
database to be an extremely large storage resource, but then, they also bring great difficulties to the routing
prediction and recognition.

3 Dynamic Combination of Genetic Algorithm and Ant Colony Algorithm

3.1 Genetic Algorithm

Genetic algorithm, which comes from Genetics and Darwin industry, is a widely used searching
method. It can be used for combinatorial optimization, pattern recognition, machine learning, planning
strategy and information processing. This algorithm has many advantages, such as: doing a global search
quickly and well; simple process of utilizing the evaluation function to search; randomness iterated by
probability mechanism; well associativity of combining with other optimization techniques.
However, It cann't make extensive use of the feedback information from the system, and would make
the blindness of search Meanwhile, when the solving progress reaches a certain rang, more redundancy
iterations are needed, then the speed of the convergency to the optimal solution will drop quickly, which
lead to solve the optimal solution inefficiently.

3.2 Ant Colony Algorithm

The ant colony algorithm based on swarm intelligence is a new bionic evolutionary algorithm
developed in recent years. It simulates the ant's foraging behaviour to solve the complicated combinatorial
optimization problem. It has a positive feedback system, and can converge the optimal solution by
continuously updating pheromones; it has global stochastic optimization characteristics, and is a distributed
optimization method, that is useful for parallel computing; it is a global optimal method, which can be used
to solve both the single-objective optimization problem and the multi-objective optimization problem[1].
Disadvantage: At the beginning, pheromones are insufficiency, so it will take too long time to seek the
accumulated initial pheromones.

3.3 Basic principle of dynamic combination between Genetic Algorithm and Ant Colony Algorithm

Through the research and experiment on genetic algorithm and ant colony algorithm, we find out a
speed-time curve of genetic algorithm and ant algorithm just like Fig.1 shows. For genetic algorithm, there
is a higher speed of the convergency to optimal solution during the preliminary stage˄t0~ta˅, but it will
significantly reduce after ta. However, during the preliminary stage˄t0~ta˅of ant algorithm, the searching
speed is very slow for lacking of pheromones, then after pheromones reach a certain degree˄after ta ˅, the
speed of convergency to optimal solution improves quickly. The basic principle of dynamic combination
between genetic algorithm and ant colony algorithm is that we can utilize genetic algorithm to get initial
 ZHANG Yan-hua et al. / Procedia Engineering 15 (2011) 3341 – 3345 3343

pheromones and then obtain the optimal solution by ant algorithm.

Dynamic integration strategy of genetic algorithm and ant colony algorithm is used to make sure
integrating at the best time.˄1˅set the minimum genetic iterations Genemin ˄for example tb˅ and the
maximum genetic iterations Genemax ˄for example tc˅.˄2˅count the evolving rate of offspring during the
genetic algorithm's iteration, and then set the minimum rate of offspring Genemin-impro-ratio according the rate
˄3˅during the set scope, if Genedie evolving rate is continuous lower than Genemin-impro-ratio, which means
the optimizing speed is too low, we can stop the genetic algorithm's iteration and begin to use the ant
colony algorithm.
v
d b Ant algorithm

v a
c e Genetic
t0 td tb t t t t
Fig.1 Speed-Time curve of genetic algorithm and ant algorithm
4 Problem description about cloud database route scheduling

Definition1, there is a connected graph to network G=˄VˈE˅, which has n nodes˄V0…n˅, here V
denotes the set of notes, E denotes the set of all sides, V denotes numbers of the node, E denotes
numbers of the side. τ (i, j ) denotes time of side (i,j), ξ (i, j ) denotes the delay form i to j. Suppose
τ (i, j ) and ξ (i, j ) are both nonnegative, we can find out the optimal paths from the appointed node V0
in graph G to other nodes. Suppose Γ(i ) is the set of i 's descendant nodes, Γ(i ) denotes numbers of
−1 −1
the descendant node, Γ (i ) denotes the set of i 's precursor nodes, Γ (i ) denotes numbers of the
precursor node. Problem of cloud database route scheduling can be classed as the shortest path problem in
the delayed network[4-6]˖
λ j ,k = min {λ jk , ξ (i, j ) + τ (i, j ) + λi , j } ˄1˅
∀i∈Γ −1 ( j )

∀j ∈ V  k ∈ Γ( j ) 
Cloud database route scheduling is a complex process involving network flow distribution,
congestion control, rights management and routing list. It will cause the delay of searching the needed
resource path from the cloud database. Routing list, the key factor of scheduling the path optimization
of cloud database system, records the algorithm needed informations (including delay informations ),
depending on which the algorithm makes the local forward decision. Unlike the conventional
centralized database and the static storage database, cloud database need a proper data storage
method, that has both distributed function and dynamic update function. In contrast, tree search
algorithm and static search algorithm are not appropriate for the cloud database. When the scale of the
cloud database is limited, deterministic algorithm can find out the optimal solution to the
combinatorial optimization problem. However, when the scale increases indefinitely, there is a bare
possibility to find out the optimal solution. The only practical way to improve the computational
efficiency is reducing the optimal solution's accuracy, and getting the approximate solution by
non-deterministic algorithm. Problem of cloud database route scheduling can be solved either in this
way. Meta-heuristics algorithm, presented by the genetic algorithm and ant colony aglorithm to solve
the NP problem and the discrete optimization problem, can solve both the static combinatorial
optimization problem and the dynamic combinatorial optimization well. It is the better conjunction
point about cloud database's dynamics Scalability and large scale[7-8]. 5 Algorithm of Cloud Database
Route Scheduling
3344 ZHANG Yan-hua et al. / Procedia Engineering 15 (2011) 3341 – 3345

5.1 Genetic algorithm rules of cloud database route scheduling

Genetic coding: binary encoding.

Object function: Cloud database route scheduling need to find the best system uptime in the constraint
conditions of network traffic distribution, congestion control, rights management and routing list, and then
the object function was defined as formula˄1˅.
Population initialization: generate the initial population by random method.
Selection Operator: utilizing the routing disc type of selection strategy which is widely used in genetic
algorithm.
Crossover Operator: utilizing the uniform crossover strategy.
Mutation Operator: reversing selected individuals by the probability of mutation pm.
Settings of GA control parameters: executing the nonoverlapping genetic process by operator; setting
population size N=80, percentage of hybrid pc =0.6, probability of mutation pm=0.2, according to the
scope of controls parameter's empirical value given by literature.
Constraint condition of genetic algorithm: In this paper, the constraint condition of genetic algorithm
actually is estimating the best fusion time of genetic algorithm and ant colony algorithm, here the author
sets the minimum genetic iterations Genemin=15, the maximum genetic iterations Genemax=60, the
minimum rate of offspring Genemin-impro-ratio=3%, Genedie=3.
5.2 Ant colony algorithm rules of cloud database route scheduling
Object function: calculating by formula˄1˅.
Pheromone setting and Update rule: Based on the pheromone setting and update rule in literature [9],
the renewal equation of the pheromone τ ij (k ) is set as:

⎧ (1 − ρ ) ∗ τ ij (k ) + Δτ ij (k ) best
⎪
τ ij (k ) = ⎨τ ij (k ) max , if τ ij (k ) > τ ij (k ) max ˄2˅
⎪τ (k ) , if τ (k ) < τ (k )
⎩ ij min ij ij min

here, ρ denotes the volatilization of pheromone. τ ij (k ) max and τ ij (k ) min denotes the maximum and
minimum of pheromone, respectively, τ ij ( k) best means the augmenter of sides' s pheromones.
For any ant, in the centre of current node, selects by the nearest and closest selection strategy and
walks to the next node. The node having the maximum probability with the current node will be the next
node ant would walk.

τ ij (k )α η j (k ) β
pij (k ) = ˄3˅
∑τ ij (l )α η j (k ) β
here τ ij (k ) denotes the edge's pheromones, α denotes the controlling elements of the global elicitation
information ,while β means the local elicitation information's controlling elements, η ij (k ) denotes the
local elicitation information.
Settings of ant colony algorithm's parameters: α = β = 1 , the volatilization of pheromone ρ =0.2.
Ant colony algorithm 's end condition: the algorithm will end when meeting one of the following
criterias:˄1˅the genetic iterations is up to Antmax,˄2˅the improvment of offspring 's optimization solution
is lower than Antmin-improv-ratio, during the continuous Antdie iteration. In this paper, Antmax=200, Antdie=5,
Antmin-improv-ratio=0.5%.
 ZHANG Yan-hua et al. / Procedia Engineering 15 (2011) 3341 – 3345 3345

6 Simulations

The fusion algorithm ,combined by the genetic algorithm and the ant colony algorithm, develops
enough advantage of the two algorithms. It can get the precision of the solution more efficiently than the
genetic algorithm and is in terms of time better than the ant colony algorithm efficiency. Therefore, it can
calculate the optimal path very well. Simulations show that the fusion algorithm do a good job on the
optimizing performance and the time performance.
In order to compare and verify the effect, the experiment, tested on the C# experimental environment,
is presented to get the running time with different numbers of computer nodes by using the ant colony
algorithm and the fusion algorithm separately. The author 's computer configuration is CPU 2.4G, RAM 2G,
hard disk 160G. The result is just like Fig.2 shows.

Fig.2 Running-Time/nodes curve

As the Fig.2 shows, there is no different between the genetic algorithm and the fusion algorithm
when the number of nodes is less than 50, but the fusion algorithm can do better than the genetic algorithm
after the number is higher than 50. Its advantage becomes more evident as the node size increases.
7 Conclusions
In this paper, the author combines the dynamic combination of genetic algorithm and ant colony
algorithm with the searching in cloud database, and good results are obtained. The bigger the nodes scale is,
the better the algorithm performs. After comparing the fusion algorithm with the ant colony algorithm, the
author obtains good simulation result. Next research direction ought to be how to optimize search
performance, minimize the search time, lower search costs and compare with other routing algorithm.
References
[1] Shi HLˈBai GYˈTang ZMˈet al. Cloud database dynamic route scheduling based on ant colony optimization algorithm[J].
˖143-145.
Computer Scienceˈ2010ˈ37˄5˅
[2] Xiong ZHˈLi SKˈChen JH. Hardware/Software partitioning based on dynamic combination
˖503-512.
of genetic algorithm and ant algorithm[J]. Journal of Softwareˈ2005ˈ16˄4˅
[3] Ding JLˈChen ZQˈYuan ZZ. On the combination of genetic algorithm and ant algorithm[J]. Journal of Computer Research and
˖1351-1356.
Developmentˈ2003ˈ40˄9˅
[4] Ioannidis YˈRamakrishman RˈWinger Lˈ Transitive closure algorithm based on graph traversal[J]. ACM Trans on Database
˖512-576.
Systemsˈ1993ˈ18˄3˅
[5] Shekar SˈKohli AˈColyle M. Path Computation algorithm for advanced traveler information systems[A]. Proc of the 9th Int’1
Conf on Data Engineering[C]. ViennaˈAustriaˈ1993˖31-39.
[6] Athanasiosk.ZiliaskopoulosˈHanis.Mahmassani. A note on least time path computation considering delays and prohibitions for
˖359-366.
intersection movements[J]. Transportation Res-Bˈ1996ˈ30˄5˅
[7] Pan ZJˈKang LSˈChen YP. Evolutionary computation[M]. Beijng˖Tsinghua University Pressˈ1998˖60-153.
[8] Flum JˈGrohe MˈThe parameterized complexity of counting problems[J]. SIAM Journal on Computingˈ2004ˈ33(4)˖892-922.
[9] Stutzle TˈHoos HH. MAX-MIN ant system[J]. Future Generation computer systemˈ2000ˈ31(8)˖1276-1280.