OPERATIONS RESEARCH AS A SERVICE

Mehdi Sheikhalishahi, Demetrio Lagan

a and Lucio Grandinetti

Department of Electronics, Computer and System Sciences, University of Calabria, Rende (CS), Italy

Keywords:

Operations Research, Cloud Computing, Online Services.

Abstract:

In this paper, we propose cloud solutions to provide Operations Research services to scientists. Currently,

users must install Operations Research software tools or develop algorithms in order to tackle their problems.

1 INTRODUCTION

Cloud computing is outsourcing of IT resources such

as software, hardware, etc. in which dynamically

scalable and often virtualized resources are provided

as a service over the Internet. The main concept in the

cloud is an infrastructure that provides on-demand,

instant and elastic resources or services over the Inter-

net, usually at the scale and reliability of a data center.

The future of software is in the cloud. It is our vi-

sion that by combining Operations Research(OR) and

cloud computing technologies, OR cloud will make a

wider audience able to easily access and beneﬁt from

the increasing number of OR services. Researchers

are familiar with OR tools, but with OR cloud we can

offer a much simpler and faster method for doing Op-

erations Research.

Our OR cloud mission is similar to Globus On-

line that is a hosted service to automate the tasks

associated with moving ﬁles between sites. It does

not require custom infrastructure, Globus Online is

software-as-a-service that can be used today without

building these features yourself. However, Operations

Research is a much more complex science than ﬁle

transfer, thus to develop a general-purpose and com-

plete solution, PaaS and IaaS cloud models should be

exploited.

Obtaining, conﬁguring, and maintaining solvers

for numerous optimization classes is an expensive

and time consuming proposition. Additionally, op-

timization problems often require signiﬁcant comput-

ing time. Running multiple problem instances, each

requiring substantial CPU resources, on a client ma-

chine such as a laptop, is not practical. The ability

to solve optimization problems in a distributed en-

vironment is an ideal solution. Lack of standards in

OR world, makes the development of OR as a ser-

vice difﬁcult. In OR world, there are numerous mod-

eling languages with their own format, diverse prob-

lem instances with their own representation, various

solvers with different APIs, results, etc. With OR in

the cloud, scientists would beneﬁt from the following

advantages: No IT Required: requires no software in-

stalls or complex IT infrastructures, Give users a tool

that simpliﬁes dealing with OR, Automates the time-

consuming and error-prone activity of doing OR such

as manual tasks, so users can stay focused on what is

most important: their research, Provide users with di-

rect access to a wide variety of OR tools and services

without the need to install, conﬁgure, or upgrade soft-

ware on her computer., Enable better collaboration

among users, Get the service up and running in min-

utes, Handling and managing of users’ dataset such

as Graph, Matrix, Network, etc., Validating and con-

verting users’ dataset into other data structures With

OR cloud, users simply specify their OR problem and

some other information e.g. which math program-

ming model they would like to use, which methods.

They will do this through a few questions. Then the

OR cloud will provision the best OR environment for

that OR problem. Next parts of this paper are orga-

nized as follows. Section 2 presents some background

information on the topic. Then, Section 3 goes over

related work. Section 4 discusses operations research

cloud solution.

2 BACKGROUND

Currently, if a researcher wants to solve an speciﬁc

problem of her research area, she would need to pass

480

Sheikhalishahi M., Laganà D. and Grandinetti L..

OPERATIONS RESEARCH AS A SERVICE.

DOI: 10.5220/0003961704800483

In Proceedings of the 2nd International Conference on Cloud Computing and Services Science (CLOSER-2012), pages 480-483

ISBN: 978-989-8565-05-1

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

through the following steps: (1) developing or ﬁnd-

ing the right and the most efﬁcient model, algorithm,

tool and software; (2) installation; (3) conﬁguration;

(4) deployment; (5) test (6) usage. In this paper, we

consider operations research and we provide cloud so-

lutions to address the aforementioned challenges in

solving OR problems for OR community.

The ﬁrst and the main obstacle in usage of soft-

ware, tools, and source code repositories is their in-

stallation and then their conﬁguration management.

Users need to install and deploy them in order to use

them. This is a big challenge for user community of

any scientiﬁc ﬁeld. The second difﬁculty of scientists

is the usage of a software and a service. In addition,

some questions such as whether a software is a good

one for a problem, whether a software is optimized for

a problem, etc. are very important for scientists in or-

der to solve their problems with the most efﬁcient and

optimal methods. In addition, installation, conﬁgura-

tion, and deployment of these OR tools, software and

services is a time-consuming task. By transforming

all these bundles in the cloud we can instantiate them

faster and users will get rid of all the difﬁcult tasks.

As a result, time to solutions will be faster than be-

fore.

There are many tools, software and services in the

market and open source software in free software to

offer appropriate solutions for various OR commu-

nity. We have diverse set of OR problems, diverse

set of models, methods, etc., etc. One solution, one

piece of software, one interface, one tools, one math

programming model does not suit the needs of all op-

erations research users. Thus, if for an OR department

we only use one OR software tool such as CPLEX,

we cannot say that all OR users will be able to use

it to solve their problems, also in the future perhaps

will be more problems which cannot be solved by

CPLEX. Therefore, we need to provide to an OR user

a suitable and customized OR environment in order to

model their problems with the most appropriate and

optimized math programming model (e.g. algebraic

lang.), and then select the best and the most efﬁcient

method to solve their problems. This OR environment

depends on the focus of research activity. In sum, the

OR cloud should be general-purpose and rich enough

in order to address the needs of all OR user commu-

nity or a large part of them.

In order to provide customized OR environment

to a user according to its need, problem, and so on,

we need to develop an OR-aware interface as an ab-

straction layer over OR tools, software, mathemati-

cal programming models, methods, algorithms, etc.

An OR-aware interface at the highest level of cloud

will determine the user needs based on a few ques-

tions and then will prepare the most efﬁcient OR en-

vironment for that user. In cases that there are more

than an option, this interface will suggest to user the

best possible options. This abstraction layer will be

implemented as part of SaaS solution of OR.

In order to implement an OR cloud system with

the aforementioned features, IaaS and SaaS cloud

paradigms should be exploited as well.

3 RELATED WORK

Globus Online (Online, ) is among the ﬁrst cloud so-

lutions for scientists, it makes robust ﬁle transfer ca-

pabilities accessible to any researcher with an Inter-

net connection and a laptop. Globus manages the en-

tire ﬁle transfer operation: monitoring performance,

retrying failed transfers, recovering from faults au-

tomatically whenever possible, and reporting status.

Users cite Globus Online as their preferred service

since it is Easy, Fast, Secure, Reliable and Research-

focused.

NEOS (Czyzyk et al., 1998) is an Online opti-

mization service which has been widely used by the

OR community for over a decade. A central server

maintains and queues job submissions for solvers that

run on a variety of workstations scattered around the

Internet. The main drawback of NEOS is its central

server paradigm.

At ﬁrst, submissions were MPS-format ﬁles for

linear problems and C or Fortran programs for non-

linear ones, but now the great majority of submis-

sions are in high-level modeling languages, pre-

dominantly AMPL (AMP, ) and GAMS (GAM,

). Submissions through the NEOS web portal

(neos.mcs.anl.gov/neos/solvers) remain popular, and

they can also be made by sending XML text ﬁles

through email. The latest NEOS release features a

NEOS API that permits all server functions to be ac-

cessed through remote function calls using the XML-

RPC protocols (www.xmlrpc.com). This has brought

NEOS more in line with the precepts of SOA and

has made it much easier to integrate into optimiza-

tion modeling environments. Nevertheless, its de-

sign still adheres in many respect to the central server

paradigm. Also NEOS employs whatever ﬁle for-

mats are supported by the various solvers; the over

40 solvers in the NEOS lineup require instance inputs

of about a dozen different kinds. Similarly there is no

NEOS standard format for communicating options to

solvers or communicating results from solvers.

In (Fourer et al., 2010), authors present a dis-

tributed optimization environment (Optimization Ser-

vices or OS) in which solvers, modeling languages,

OPERATIONSRESEARCHASASERVICE

481

Table 1: Traditional OR solution vs. its corresponding OR cloud solution.

Traditional OR solution Provisioning actions and time Cloud OR solution Provisioning actions and time

Developer tools I.C.D.T.B long IaaS, SaaS B short

Documentation B short SaaS:OR-aware cloud interface B short

Graphs I.C.D.T.B long IaaS, PaaS B short

Interfaces I.C.D.T.B long IaaS, PaaS, SaaS B short

Modeling systems I.C.D.T.B long IaaS, PaaS B short

Metaheuristics I.C.D.T.B long IaaS, PaaS B short

Optimization utility I.C.D.T.B long IaaS, PaaS B short

registries, analyzers, and simulation engines are im-

plemented as services. It deﬁnes standards for decen-

tralized optimization on the Internet: representation

of optimization instances, results, and solver options;

communication between clients and solvers; and dis-

covery and registration of optimization-related soft-

ware using Web Services. The OS project addresses

NEOS weaknesses.

In all, while OS provides online distributed op-

timization services as a SaaS cloud, but it is not a

complete cloud solution. Users will not be able to

build their own OR applications if they cannot be of-

fered by SaaS solution. In order to approach a more

complete solution, IaaS and PaaS cloud paradigms are

suggested to be used in this paper.

4 OPERATIONS RESEARCH

CLOUD SOLUTIONS

In the previous section we enumerated various OR

projects, tools, software packages, libraries, etc. in

OR world. We would like to reuse and to exploit al-

ready available solutions in OR world with the help

of IaaS, PaaS and SaaS cloud paradigms to build OR

cloud solutions. SaaS paradigm suits to reuse soft-

ware without signiﬁcant changes such as tuning con-

ﬁguration parameters, while when we want to build

new OR applications based on available OR solutions

and our development, IaaS and PaaS paradigms are

the right choice, for example to link a model to an in-

terface and exploit our speciﬁc OR algorithm for an

speciﬁc OR problem.

Table 1 provides some information on the cor-

responding cloud solution for each category of the

aforementioned OR projects in the previous section,

and compare the provisioning time and actions of tra-

ditional OR projects vs. OR cloud solutions.

In addition to free software resources, there are

many enterprise OR tools such as CPLEX that can be

hosted in the cloud. With these rich set of OR re-

sources that are hosted in the cloud, OR cloud will be

able to intelligently offer OR services for OR commu-

nity. In the provisioning column we use the ﬁrst letter

of provisioning actions to represent them i.e. I. for

installation, C. for conﬁguration, D. for deployment,

T. for test, and B. for boot. This table 1 demonstrates

applicability of cloud paradigms to OR world. Unlike

traditional OR solution with this, OR world would be

available online without signiﬁcant attempt to provi-

sion OR environment.

Figure 1 presents a high level architecture of OR

cloud. OR-aware interface is an abstraction layer over

OR tools, software, mathematical programming mod-

els, methods, algorithms, etc. An OR-aware interface

at the highest level of cloud will determine the user

needs based on a few questions and then will provide

the most efﬁcient OR environment for that user, like

in (Fourer et al., 2010) that Optimization problems

are solved automatically with minimal input from the

user. Users only need a deﬁnition of the optimization

problem; all additional information required by the

optimization solver is determined automatically. In

cases that there are more than an option, this interface

will suggest to user the best possible options. This

abstraction layer will be implemented as part of SaaS

solution of OR. In (Fourer et al., 2010), if users are not

sure of the type of optimization problem, they should

consult the Optimization Tree of the NEOS Guide for

information on optimization problems. However, we

move this action to the cloud i.e. OR-aware interface

layer. The choice of solver is then dictated by the lan-

guage used to deﬁne the optimization problem.

At IaaS we have a number of Virtual Appliances

abstracted as Virtual Machines (VM). IaaS VMs are

developed by reusing available OR tools. They are

prepared by OR experts of various ﬁelds of OR.

These VMs contain customized OR environments for

different OR problems. We reuse OR tools, soft-

wares, services, open source software, etc. in order

to build these VMs. With development environment

(Platform-as-a-Service) on the cloud, users will be

able to develop their own OR applications by exploit-

ing already available OR tools, algorithms, methods

in the cloud. Like any cloud effort, standardization

at the interface and data levels are very important, for

example open standards for data interchange.

CLOSER2012-2ndInternationalConferenceonCloudComputingandServicesScience

482

Figure 1: Operations Research cloud architecture.

REFERENCES

Ampl. http://www.ampl.com.

Gams. http://www.gams.com.

Czyzyk, J., Mesnier, M. P., and Mor

e, J. J. (1998). The neos

server. IEEE Comput. Sci. Eng., 5:68–75.

Fourer, R., Ma, J., and Martin, R. K. (2010). Optimization

services: A framework for distributed optimization.

Operations Research, 58(6):1624–1636.

Online, G. Globus online project. http://

www.globusonline.org/.

OPERATIONSRESEARCHASASERVICE

483