multi agent approach for distributed information systems reliability prediction

Kotov Tula State University, 92, Lenin Ave., Tula, 300012, Russia Abstract Multi-Agent algorithms of obtaining the rational solution of problems of ensuring stability of functioning o

Procedia Computer Science 103 ( 2017 ) 416 – 420

1877-0509 © 2017 Published by Elsevier B.V This is an open access article under the CC BY-NC-ND license

( http://creativecommons.org/licenses/by-nc-nd/4.0/ ).

Peer-review under responsibility of the scientific committee of the XIIth International Symposium “Intelligent Systems”

doi: 10.1016/j.procs.2017.01.003

ScienceDirect

XIIth International Symposium «Intelligent Systems», INTELS’16, 5-7 October 2016, Moscow,

Russia Multi-agent approach for distributed information systems reliability

prediction D.O Yesikov*, A.N Ivutin, E.V Larkin, V.V Kotov

Tula State University, 92, Lenin Ave., Tula, 300012, Russia

Abstract

Multi-Agent algorithms of obtaining the rational solution of problems of ensuring stability of functioning of the distributed information systems in the conditions of rigid temporary restrictions are offered The short characteristic of the offered algorithms

is given For neutralization of initial initialization of algorithms it is offered to use the island scheme of the organizati on of search

of the decision in the algorithms considered the Multi-Agent

© 2017 The Authors Published by Elsevier B.V

Peer-review under responsibility of the scientific committee of the XIIth International Symposium «Intelligent Systems»

Keywords: algorithm of the school of fishes; multi-agent algorithms;distributed information system; genetic algorithm; method of a swarm of

particles

1 Introduction

Reliable and smooth functioning of the distributed information systems is guarantee of effective functioning of the organizations and entities in various spheres of economy Stability of functioning of system is understood as its capability to carry out the assigned functions with the set quality indicators in the conditions of impact of the internal and external destabilizing factors

Now the most effective for ensuring stability of functioning of the distributed information systems (DIS) are the following approaches 1-4:

* Corresponding author

E-mail address: mcgeen4@gmail.com

© 2017 Published by Elsevier B.V This is an open access article under the CC BY-NC-ND license

( http://creativecommons.org/licenses/by-nc-nd/4.0/ ).

Peer-review under responsibility of the scientific committee of the XIIth International Symposium “Intelligent Systems”

1 Enhancement of information process, including regarding ensuring its stability;

2 Creation of highly reliable subsystems of reservation of data

For the rational organization of information process it is reasonable to perform storage and operation of the software elements (SE) and information massifs (IM) in storage centres and information processing (SCIP)

For ensuring stability of functioning of the distributed information systems it is offered to solve the following problems 1-3:

1 Optimization of distribution of elements of the software of functional tasks on network nodes;

2 Optimization of distribution of information resources on storage centers and data processing;

3 Determinations of the rational expense level on forming of a complex of means of data storage in storage centers and information processing;

4 Optimization of structure of technical means of a storage system and data processing;

5 Optimization of distribution of an allowance of information resources on storage centers and data processing These tasks belong to the class of tasks of combinatorial optimization 1,4, namely integer discrete programming with Boolean variables 1,3,4 and differ in a large number of variables and restrictions Besides, some of them have restrictions of a non-linear look that complicates application of traditional methods for their decision

For obtaining the exact solution of tasks of support of stability of functioning of the distributed information systems there is rather large number of methods of the discrete optimization 2,3 However with increase in dimensionality of tasks decision time by exact methods increases exponentially that doesn't allow to receive quickly their decision at an appropriate level to corporate In these conditions and justified obtaining the quasioptimal (rational) solution in time not exceeding the given is expedient The method of obtaining the rational solution of tasks of support of stability of functioning of the distributed information systems shall have adequate accuracy, low computing complexity, nonsensitivity to a type of target function and restrictions The group the Multi-Agent of algorithms and methods of peephole optimization on the basis of stochastic search meets all these requirements 5 The most well studied from this group of methods and algorithms, applicable for the solution of tasks of support of stability of functioning of the distributed information systems are:

• method of a swarm of particles (MSP) 6 - 10;

• genetic algorithm (GA) 11 - 12;

• algorithm of the school of fishes (ASF) 13, 14

In case of creation of evolutionary and population algorithms for the organization of search of the decision use in many respects similar concepts and mechanisms At the expense of a reasonable choice of parameters of algorithm, unlike the existing approximate methods, provides the guaranteed obtaining the rational (quasioptimum) decision (a series of rational decisions) in time not exceeding given

In case of the solution of tasks of support of stability of functioning of the distributed information systems two diagrams of generation of individuals for formation of the initial populations were considered:

1 Accidental generation of a candidate solution (individual) with the subsequent check on feasibility of restrictions

2 Sequential generation of a candidate solution (individual) with step-by-step check of feasibility of restrictions Both diagrams are based on accidental generation of values of variables in a task candidate solution about a satchel The first diagram potentially has more chances to create already in initial population the optimal solution, however requires essential elimination of the options which aren't satisfying to restrictions

According to the second scheme if at the subsequent stage the received values of variables in version of the solution

of a task on a satchel do not satisfy to restrictions, then the individual received at the previous stage undertakes version

of the decision This scheme forms initial population for the quantity of steps which is slightly exceeding the population size

In genetic algorithm use the following populations of individuals:

• initial population Serves for storage of the individuals participating in the current step of GA On an initial step of

GA initial population is usually generated in a random way;

• parental population (parental pool) In parental population all individuals of initial population whose value of function of fitness there is not less average value of function of fitness for initial population in general are selected

• elite population Serves for preservation of the best individuals of parental (initial) population

• affiliated population It is formed by repeated performance of operations of selection of individuals, a crossing over,

a mutation On the basis of affiliated population initial population for the following step of GA is formed

2 The island genetic algorithm

The island multi-agent algorithm with migration of individuals with topology of a ring consists of the following steps:

1 For each k-island (k=1,2, ,N Lnd) do:

2 Initialization of initial population by filling with her individuals generated in a random way Initial population doesn't join impractical individuals Repeated inclusion of already available individual isn't allowed; formation of elite

population volume L el individuals

3 i = 0

4 For each island (k=1,2, ,N Lnd ) consecutive evolution of generations in quantity is carried out N gen

(performance of simple mnogoagentny algorithm with the set scheme of a reproduction)

5 If N Lnd >1, for each k-island (k=1,2, , N Lnd -1) transfer is carried out m (m= L el /2) the best individuals from

elite population in initial population (j=k+1) j-island by replacement of the worst individuals For a case k = N Lnd

transfer of the best individuals is carried out to the 1st island

6 i = i+1

7 If i < N Step go to step 3 Otherwise to claim 7

8 To form the resultant population of individuals that make up the elite of the population of each of the islands

9 From the resulting population to choose the best individual - the result of decisions

10 The end

3 The experimental verification of the use of multi-agent algorithms

Due to the discretization of values of variable solvable tasks in algorithms of MSP and ASF it is necessary to execute operations of adaptation of retrieval procedures to a binary type of values of variables 15

In a figure 1 dependence of value of target function of the solvable task (dimensionality 150х4) from number of iterations is provided From a figure 1 it is visible that the provided algorithms realize successive approximation to a global extremum At the same time, the IGA has advantage as the received decision The simple genetic algorithm, a method of a swarm of particles and algorithm of a school of fishes provide several worst decision At the same time, since some iteration, the speed of a gain of quality of the received decision significantly decreases It demonstrates that in case of a rigid limit of time allowed for the solution of the task when it is possible to be satisfied with the rational (quasioptimum) decision there is enough use of 4-5 iterations of the selected algorithm that will allow to reduce search time of the decision on 40-60 at least %%

In Figure 2 results of an assessment of dependence of influence of accuracy of the received decision for the general time of the solution of a task of IGA of dimension 100х4 are presented Change of accuracy of the received decision was carried out by a variation of values of parameters of algorithm For each option of the IGA parameters the problem

of each dimension was solved on 10 times

From figure 2 it is visible that reduction of time of obtaining the rational decision by 3-6 times in comparison with time of obtaining the optimal solution leads to deterioration of the received solution of everything for 0.5% of optimum At the same time, with growth of dimension of a solvable task this effect becomes more obvious This circumstance allows to draw a conclusion on prospects and expediency of application of the island scheme of the organization of calculations (both in IGA, and in MSP and ASF) for expeditious obtaining the rational (quasioptimum) decision in the conditions of rigid temporary restrictions

Fig 1 Dependence of value of criterion function on iteration (task 150х4)

Fig 2 Assessment of influence of accuracy of the decision for the period of the solution of a task 150х4

(T * - time of obtaining the optimal solution)

4 Conclusion

Thus multi-agent algorithms to rapidly get a quasi-optimal solution of tasks of ensuring the sustainability of distributed information systems, and with the use of schemes of the island to find a solution have good adaptability to parallelization, which will provide the quasi-optimal solutions in less time for large-scale problems

2950 3050 3150 3250 3350 3450

1 2 3 4 5 6 7 8 9 10

W

iteration

genetic algorithm

optimal solution island genetic algorithm

References

1 Yesikov DO., Yesikov OV, Savvateev SS, Strukov AV, Sukharev EM Mathematical models, methods and algorithms to ensure the sustainability

of modern storage systems and data processing International Science -Technical journal "Nonlinear world", Moscow, JSC "" Radio

"Publishers, 2013, pp 621-632

2 Yesikov DO, Ivutin AN Complex of Mathematical Models to Ensuring Sustainability of the Distributed Information Systems 2015 4 th

Mediterranean Conference on Embedded Computing (MECO), Budva, Montenegro, June 14 th – 18 th 2015 pp 106-110 (SCOPUS)

3 Yesikov DO challenges of sustainable functioning of distributed information systems International scientific-practical journal "Software and Systems", № 4 (112), 2015, Tver Research Institute "Centerprogrammsystem"

4 Karpenko AP Population algorithms global search engine optimization Overview of new and little-known algorithms Information Technology

2012 №7 with 13-15

5 Poli, R An analysis of publications on particle swarm optimisation applications Technical Report CSM-469 (Department of Computer Science, University of Essex, UK) 2007

6 Poli, R Analysis of the publications on the applications of particle swarm optimisation Journal of Artificial Evolution and Applications:

1-10 DOI:1-10.1155/2008/685175 2008

7 Karpenko AP, Seliverstov EY Review particle swarm optimization for the problem of global optimization (Particle Swarm Optimization) Science and education 2009 Vol 03 URL: http://technomag.edu.ru/doc/116072.html (reference date: 05/01/2013)

8 Eberhart RC, Kennedy J Particle swarm optimization Proc IEEE Intern Conf on Neural Networks Piscataway (NJ): IEEE Service Center,

1995 4 pp 1942–1948 50

9 Clerc M Particle swarm optimization Hoboken (NJ): Wiley-Interscience, 2006 243 p

10 Gladkov LA Genetic algorithms / Gladkov LA, Kureichik VV, Kureichik VM M.: FIZMATLIT, 2006 320 p Holland J H Adaptation in Natural and Artificial Systems: 2nd ed Cambridge, MA: MIT Press, 1992

11 Bastos-Filho CJA., Lima-Neto FB, Lins A, Nascimento A, Lima M Fish school seach Nature-Inspired Algorithms for Optimisation SCI Springer, Heidelberg, 2009.Vol.193 pp 261-277

12 Cavalcanti-J’unior GM, Bastos-Filho CJA, Lima-Neto FB, Castro R.M.C.S AHybrid Algorithm Based on Fish School Searchand Particle Swarm Intelligence 2011 (ICSI).V.2 pp 543-552

13 Yesikov DO, Ivutin AN Evaluating the effectiveness of the reproduction of the adaptive scheme in the island genetic algorith m for solving the problems of ensuring the sustainability of distributed information systems News of the Tula State University, engineering sciences, Issue 9, Tula, Ed Tula State University, 2015 pp 119-128

14 Yesikov DO, Ivutin AN Evaluating the effectiveness of the reproduction of the adaptive scheme in the island genetic algorith m for solving the problems of ensuring the sustainability of distributed information systems News of the Tula State University, engineering sciences, Issue 9, Tula, Ed Tula State University, 2015 pp 119-128

15 Akinshin ON, Yesikov DO, Akinshina NY Feature of the solution of a task of optimization of an investment portfolio of the entity by method

of a swarm of particles News of the Tula State University, engineering sciences, Issue 5, Tula, Ed Tula State University, 2016 pp 109-116