Design of a New Cloud Computing Simulation Platform - Springer Link

Design of a New Cloud Computing Simulation Platform A. Nuñez1 , J.L. Vázquez-Poletti2, A. C. Caminero3 , J. Carretero1 , and I. M. Llorente1 1 Dep. de Informática Universidad Carlos III de Madrid. Spain {anunez,jcarretero}@inf.uc3m.es 2 Dept. de Arquitectura de Computadores y Automática Universidad Complutense de Madrid. Spain [email protected] 3 Dept. de Sistemas de Comunicación y Control Universidad Nacional de Educación a Distancia. Spain [email protected], [email protected]

Abstract. Cloud computing is a paradigm which allows the use of outsourced infrastructures in a “pay-as-you-go” basis, thanks to which scalable and customizable infrastructures can be built on demand. The ability to infer the number and type of the Virtual Machines (VM) needed determines the final budget, thus it represents a key in order to efficiently manage a cloud infrastructure. In order to develop new proposals aimed at different topics related to cloud computing (for example, datacenter management, or provision of resources), a lot of work and money is required to set up an adequately sized testbed including different datacenters from different organizations and public cloud providers. Therefore, it is easier to use simulation as a tool for studying complex scenarios. With this in mind, this paper introduces iCanCloud, a novel simulator of cloud infrastructures with remarkable features such as usability, flexibility, performance and scalability. This tool is specially aimed at simulating instance types provided by Amazon, so models of these are included in the simulation framework. Accuracy experiments conducted by means of comparing results obtained using iCanCloud and a validated mathematical model of Amazon in the context of a given application are also presented. These illustrate the efficiency of iCanCloud at reproducing the behavior of Amazon instance types. Keywords: Cloud computing, simulations, validation, flexibility, scalability.

1 Introduction Cloud computing [1] [2] is a paradigm which provides access to a flexible and ondemand computing infrastructure, by allowing the user to start a required number of virtual machines (VM) to solve a given computational problem. If the same software and configurations are needed, the VMs may be started using the same image. This way,

This research was supported by the following projects: Spanish Ministry of Science and Innovation under the grant TIN2010-16497, MEDIANET (Comunidad de Madrid S2009/TIC1468) and HPCcloud (MICINN TIN2009-07146).

B. Murgante et al. (Eds.): ICCSA 2011, Part III, LNCS 6784, pp. 582–593, 2011. c Springer-Verlag Berlin Heidelberg 2011

Design of a New Cloud Computing Simulation Platform

583

a machine offered by a cloud environment may become whatever the user needs, from a standalone computer to a cluster or grid node. As soon as the scientific community had access to cloud production infrastructures, the first applications started to run on the cloud [3] [4]. In many research areas, the leap from traditional cluster and grid computing to this new paradigm has been mandatory, being the main reason an evolution in the computational needs of the applications [1]. A remarkable fact from this evolution is that in a pre-cloud environment, hardware defines the level of parallelism of an application. In cloud computing, the level of parallelism is defined by the application itself, as there are no restrictions in terms of number of machines, and CPU availability is 100% guaranteed by standard. There are mainly two cloud infrastructure types. On the one hand, a private cloud is a system where the user’s institution maintains the physical infrastructure where the VMs will be executed. These cloud infrastructures can be built using virtualization technologies like Nimbus [5], OpenNebula [6] or Eucalyptus [7]. On the other hand, the cloud service can be outsourced by paying each deployed VM per unit of time basis - this being called public cloud. Some examples of public clouds are ElasticHosts 1 and Amazon’s Elastic Compute Cloud 2 . In order to develop new proposals aimed at different topics related to the clouds (for example, datacenter management [8], or provision of resources [9]), a lot of work and money is required to set up an adequately-sized testbed including different datacenters from different organizations and public cloud providers. Even if automated tools exist to do this work, it still would be very difficult to produce a performance evaluation in terms of time and budget, due to the great number of possible setups that a typical cloud infrastructure provides. Therefore, it is easier to use simulation as a way to study complex scenarios. This paper introduces iCanCloud, a simulator of cloud systems. This tool is specially aimed at simulating instance types provided by Amazon, thus models of these are included in the simulation framework. The main contributions of this paper are: (1) the development of iCanCloud, a simulator for cloud systems; and (2) the validation of the simulator, which has been carried out by comparing the results from a validated mathematical model of instance types provided by Amazon in the context of a given application with the model of the same instances using iCanCloud. The rest of the paper is structured as follows: Section 2 presents related work in simulations in computer science; Section 3 details the architecture of iCanCloud, its most important features, and depicts the structure of the simulations executed using it; Section 4 presents the accuracy experiments, mentioned above. Finally, Section 5 draws conclusions and suggests guidelines for future research.

2 Simulators in Computer Science Simulations have been widely used in different fields of computer science over the years. For instance, in the networking research area we can find NS-2 [10], DaSSF [11], OMNET++ [12], OPNET [13], and J-Sim [14], among other simulation tools. These 1 2

http://www.elastichosts.com/ http://aws.amazon.com/ec2/

584

A. Nuñez et al.

simulators are focused on network details, such as network protocols, path discovery, latencies, or IP fragmentation, but lack the details to simulate virtualization-enabled computing resources and applications. The work [15] presents a simulation platform for modeling HPC architectures called SIMCAN. This platform is aimed at testing both existent and new designs of HPC architectures and applications. SIMCAN employs a modular design that eases the integration of the different systems on a single architecture. The design follows an hierarchical schema that includes simple modules, basic systems (computing, memory managing, I/O and networking), physical components (nodes, switches, ...) and aggregations of components. Works sharing the same aim as SIMCAN are GEMS [16] and SimFlex [17], among others. For Grids, another set of simulators have been developed, such as GridSim [18], OptorSim [19], SimGrid [20] and MicroGrid [21], among others. These tools can simulate brokerage of resources, or execution of different types of applications on different types of computing resources, but as before they lack the details to simulate a cloud environment. Focusing on Cloud Computing, [22] introduces a model for characterizing the usage of resources pertaining to a private cloud infrastructure. However and to the authors’ knowledge, the only complete tool that can simulate a real cloud system is CloudSim [23]. Although CloudSim was presented very recently [24], several research articles have been published presenting results obtained with it [9] [25] [26] [8]. This tool was initially based on a grid simulator [24] (this being GridSim [18]). So, a new layer on top of GridSim was implemented to add the hability to simulate clouds. But the first versions of CloudSim presented many bugs, and in-depth re-implementations took place to fix this. These re-implementations include a full implementation of the SimJava simulation kernel, which was the root of many of the problems of CloudSim. CloudSim has been re-designed from scratch so that it does not rely on GridSim any more. Thanks to this, most of the problems of the simulator were fixed.

3 iCanCloud Simulator The ever-increasing complexity of computing systems has made simulators a very important choice for designing and analyzing large and complex architectures. In the field of cloud computing, simulators become specially useful for calculating the trade-offs between cost and performance in “pay-as-you-go” environments. Hence, this work describes a simulation platform for modeling and simulating large cloud environments, which represent both actual and non-existent cloud computing architectures. The main aim of this tool, called iCanCloud, is to predict the trade-offs between cost and performance of a given application executed in a specific hardware, and then provide users with useful information about such costs. iCanCloud can be used by a wide range of users, from basic active users to developers of large distributed applications or system administrators. The iCancloud framework provides a scalable, flexible, fast and easy-to-use tool which lets users obtain results quickly in order to help them to make a decision regarding both the number and type of machines to use – which clearly affects the budget of


585

the user. Therefore, it provides a set of components that allow to create cloud computing scenarios easily; these components represent the behavior of actual components that belong to actual architectures, like disks, networks, memories, file systems, etc. Thus, those components are hierarchically organized in the repository of iCanCloud, which makes up the core simulation engine. Apart from designing simulated environments using built-in components provided by iCanCloud, new components can be added to its repository. Moreover, iCanCloud allows an easy substitution of components for a particular feature (e.g. different network adaptors can be switched easily). Those interchangeable components can differ in level of detail (to allow performance versus accuracy trade-offs), in the functional behavior of the component, or both. This repository of components is an interesting feature that helps to make iCanCloud a versatile simulation tool. 3.1 Features The most remarkable features of the iCanCloud simulation platform are: – Both existing and non-existing cloud computing architectures can be modeled and simulated. – Customizable VMs can simulate easily both uni-core/multi-core systems using several scheduling policies. – The memory, storage, and network subsystems can be modeled for simulating a wide range of real systems. – Network system can be modeled for simulating a wide range of distributed environments with a high level of detail. – iCanCloud provides a user-friendly GUI to ease the generation and customization of large distributed models. This GUI is specially useful for: • Managing a repository of pre-configured VMs. Thus, a set of VMs can be fully customized in order to be used for building cloud computing systems. • Managing a repository of pre-configured cloud systems. Thus, each cloud system can be built using the pre-configured VMs from the repository and also establishing a cost policy for each system. • Managing a repository of pre-configured experiments. Thus, each simulated environment can be managed from this GUI, maintaining a collection of simulations with its corresponding results. • Launching experiments from the GUI, which ease users to start simulation without using a command line console. However, command-line executions are also supported. • Generating graphical reports (in pdf format), which lets users understand easily the results obtained from simulations. – iCanCloud provides an API for developers, whereof new application models can be included easily. – New components can be added to the repository of iCanCloud, increasing the functionality of the simulation platform.

586

A. Nuñez et al.

Cloud system

Rent policies ($/h)

Vms architecture

Cloud Broker Job management

Application Repository Software models

Phobos

Local application

Typical HPC model

Map-reduce

Cost/Performance Heuristics

Cost policies

...

Single Local Application model

API

Vms Repository

Storage system

General Purpose Hard Disks

Medium Small - 2 CPU core - 1 CPU core - 1 Disk - 1 Disk - 2 GB Ram - 1 GB Ram ... Hardware ...

CPU system

Memory modules Single-core Processor

...

Network System Network Network interfaces interfaces

Multi-core Processor

Solid State Disks

...

Memory system

High-CPU - 4 CPU core - 4 Disks - 16 GB Ram ...

...

...

Communication Communication devices devices Communication links

...

Fig. 1. Basic layered schema of iCanCloud architecture

This set of features allow the development of simulations of real cloud systems. Thus, researchers and practicioners can easily implement models of their systems of interest, which permits them analyzing their systems more efficiently than if a real system had to be implemented each time. 3.2 Architecture of iCanCloud iCanCloud has been developed on top of the SIMCAN simulation framework [15]. Thus, the models of real hardware components have been used for creating the core engine of iCanCloud. The basic idea of a cloud system is to provide users with a pseudo-customizable infrastructure where they can execute specific software. The architecture of iCanCloud has been designed based on this principle in order to model full cloud infrastructures, on top of which other services can be built and deployed, from a single application (Software as a Service, SaaS) to a development platform (Platform as a Service, PaaS). Figure 1 shows the layered architecture of iCanCloud. This simulation platform can be split in two different sections. On the one hand, the section that provides components for modeling and simulating hardware and software elements: the iCanCloud core engine (dark grey). On the other hand is the section that contains those models for configuring the cloud system, like VMs, user’s jobs, and cost policies, which must be defined by users (light grey). The bottom of the architecture consists of the hardware layer. This layer is in charge of modeling the physical parts of a system, like disk drives, memory modules, communication networks and CPUs. Using those models, entire distributed systems can be modeled


587

and simulated. In turn, this section consists of four groups, where each corresponds to a specific basic system: storage, processing (CPU), memory and network systems. The upper layer is a repository of VMs, and contains a collection of VMs previously defined by the user. Initially, the iCanCloud simulator provides several models of VMs that exist in the well known public cloud from Amazon EC2. Moreover, users can add, edit, or remove VMs from this repository. In a cloud system, the VM is the most relevant component. Similarly, in iCanCloud a VM is a building block for creating cloud systems as explained before. The key of this simulation platform is its modularity, which allows to create complex modules by using other modules previously defined. Thence, the basic idea of iCanCloud consists on using VMs modules for building entire cloud computing systems. In a cloud, virtual machines are in charge of hiding the hardware details, providing to the users a logic view that corresponds with the user requirements. Thus, the VM models defined in this layer use the hardware components defined in the lower layer. Next layer, called software models, contains the application models provided by the iCanCloud core engine. These models are used for modeling the behavior of a wide spectrum of applications. The basic idea is to let users customize those models for creating a specific application models, which are executed in specific environments defined by a set of VMs. In the current version of this simulator there are three generic models for modeling applications. Each one of those models can be fully customized by setting a specific set of parameters by the user. Moreover, new application models can be easily added to the system, because iCanCloud provides an API in order to ease the development of new application models. This API contains a set of functions for using the four previously described systems for the hardware layer. The application repository layer contains a collection of pre-defined applications customized by users. Similarly to the repository of VMs, initially this repository provides a set of pre-defined application models. Those models will be used in order to configure the corresponding jobs that will be executed in a specific instance of a VM in the system. The layer on top, called cloud broker, consists on a module in charge of managing all incoming jobs and the instances of VMs where such jobs will be executed. When a job finishes its execution, this module is in charge of releasing the VMs that are currently idle, and then re-assigning the available resources in the system to execute the remaining jobs. This module also contains cost policies in order to assign incoming jobs, and then, depending on the policy selected, jobs will be assigned to a specific instance chosen by the corresponding heuristic. Finally, at the top of the architecture is the cloud system layer, which contains a definition of a set of the VMs that composes the entire cloud system and a definition of cost policies.

4 Accuracy Experiments After a simulator has been developed, implemented, and debugged, it must be tested for correctness and accuracy. Performance model validation involves generating test cases, simulating the model under test, and comparing execution results to a known reference.

588

A. Nuñez et al. Table 1. Characteristics of different machine types offered by Amazon EC2 Machine Type Cores C.U. Memory Platform Price/h Standard On-Demand Instances Small (Default)

1

1

1.7GB

32bit $0.085

Large

2

2

7.5GB

64bit $0.34

Extra Large

4

2

15GB

64bit $0.68

High CPU On-Demand Instances Medium

2 2.5

1.7GB

32bit $0.17

Extra Large

8 2.5

7GB

64bit $0.68

Model validation is usually defined to mean “substantiation that a computerized model within its domain of applicability possesses a satisfactory range of accuracy consistent with the intended application of the model” [27]. However, determining that a simulator is absolutely valid over the complete domain of its whole intended field of applicability is a very hard and time-consuming task. Thus, the level of accuracy of a given simulator cannot be calculated for the entire domain this simulator is targeted, because this accuracy depends directly on the system to be modeled. In this paper a validation process has been conducted to demonstrate the applicability and usefulness of the iCanCloud simulator. This process consists on comparing the results obtained from a validated mathematical model of the Phobos application [28], with results from the analogous model using iCanCloud. This model consists of the simulation of the application, and the corresponding hardware environment where that application has been executed. The application chosen for this validation calculates the trajectories of Phobos, the Martian moon, in the context of the Finnish-Russian-Spanish Mission to Mars that will be launched in 2011 [29], which was ported to the Amazon EC2 public cloud infrastructure [28]. Pertaining to the parameter sweep execution profile, the resulting application divides the overall tracing interval in subintervals that are calculated by the subsequent tasks in the cloud – thus the tracing interval of a task is not related to its execution time. The tracing interval processed by each task is the same and the system performs dynamic scheduling where a continuous polling of free cores guarantees a constant resource use. As was explained before, the chosen cloud infrastructure is Amazon EC2, which has become the de facto standard public cloud infrastructure for many scientific applications. The baremetal infrastructure providing the services is located in two locations in USA, one in Asia and another one in Europe. The users may choose from a wide range of machine images that can be booted in one of the offered instance types. Depending the chosen instance type, the number of CPU core number, core speed, memory and architecture differ, as shown in Table 1. The speed per CPU core is measured in EC2 Compute Units, being each C.U. equivalent to a 1.0-1.2 GHz 2007 Opteron or 2007 Xeon processor. Nevertheless, accessing to an almost infinite computing infrastructure has its price, which depends on the instantiated VM type per hour, also shown in the same Table.


589

The paper which describes the porting of the Phobos tracing application [28] introduced and validated an execution model, along with a study of the best infrastructure setup by means of instance types and number. In order to deal with the complexity level added by an infrastructure following a pay-as-you-go basis, a metric named Cost per Performance (C/P ) was also provided: C/P = CT =

Ch Texe I Texe I iNc2 iNvm Nc

(1)

where T exe is the task execution time, the values of I and i correspond to the whole tracing interval and the tracing interval per task, that is, the grain of the application. On the other hand, Nvm and Nc are the number of Virtual Machines and number of cores per Virtual Machine, as shown in Table 1 along with the machine’s usage price per hour (Ch ). This way, the best infrastructure setup would be that which produced the lowest C/P value. Figures 2 and 3 present results of executions of the model of the Phobos application along with the results of the same application implemented on iCanCloud. Each figure represents the C/P metric for the experiments, where the Small and High CPU Medium 0.5 interval years 1 interval year

1400

2 interval years

C/P Metric

1200

3 interval years

1000

4 interval years

800

5 interval years 6 interval years

600

7 interval years

400

8 interval years

200 0 5

10

15

20

25

30

35

40

Number of Virtual Machines

(a) Mathematical model 0.5 interval years

1400

1 interval year 2 interval years

C/P Metric

1200

3 interval years

1000

4 interval years

800

5 interval years 6 interval years

600

7 interval years

400

8 interval years

200 0 5

10

15

20

25

30

35

40


(b) iCanCloud Fig. 2. Simulation of Phobos using a Small instance (model and iCancloud)

590

A. Nuñez et al.

0.5 interval years

160

1 interval year

C/P Metric

140

2 interval years

120

3 interval years

100

4 interval years

80

5 interval years

60 40 20 0 5

10

15

20

25

30

35

40


(a) Mathematical model 0.5 interval years

160

1 interval year

140

2 interval years

C/P Metric

120

3 interval years

100

4 interval years

80

5 interval years

60 40 20 0 5

10

15

20

25

30

35

40


(b) iCanCloud Fig. 3. Simulation of Phobos using a High CPU Medium instance (model and iCancloud).

instance types provided by Amazon are used, and number of VMs and tracing intervals are varied. Mainly, the most relevant difference between the iCanCloud and the mathematical model is the variations obtained when the number of VMs increases. In the results obtained from iCancloud, we can see that in some cases, using the same size for the interval (in years) and increasing the number of VMs, causes an increase in the C/P metric, which does not happen in experiments using the mathematical model of Amazon. It is mainly caused because the cost of the each VM is measured in completed hours, where an hour cannot be split in fractions. Then, increasing the number of VMs provides the same execution time, increasing the cost for this configuration. Logically, the greater number of VMs used, the greater cost of the system. This effect only appears when the number of VMs gets higher. When the number of VMs is low, the performance gain when more VMs are used justifies the increase in the cost. However, the overall system performance obtained using the Amazon schema is reflected in the model using iCanCloud, which is the main goal pursued by this simulation platform.


591

5 Conclusions and Future Work In this paper, a simulator of cloud systems, called iCanCloud is presented. Its main features are highlighted, along with details of its inner architecture. Among others, iCanCloud is versatile, flexible, and scalable. iCanCloud is specially aimed at simulating instance types provided by Amazon, so models of these are included in the simulation framework. This paper presents a validation of iCanCloud, in which it is compared with a validated mathematical model of Amazon in the context of a given application, which illustrates its ability to simulate actual Amazon EC2 instance types. This validation has been conducted using an application of the astronomy domain which calculates the trajectories of Phobos, the Martian moon, over a tracing interval. This is done by dividing the overall tracing interval in identical subintervals, each of them executed by a different task. Results show that iCanCloud produces similar results to the validated mathematical model in terms of Cost per Performance (C/P ), thus iCancloud is a valid tool to similate Amazon EC2 instance types. Regarding lines for future work, it will be interesting to extend iCanCloud’s support to other providers in order to perform a budget and performance study depending on the desired infrastructure setups and/or applications. Thanks to iCanCloud’s modularity, it will be possible to recreate a hardware setup in a hardware infrastructure. With this in mind, another future step is to extend the simulator to private clouds, simulating the behaviors of the different virtual infrastructure managers available. In order to enhace the scalability of iCanCloud, another interesting guideline is making the simulator parallel. This way, one single experiment can be executed spanning more than one machine, which allows larger experiments to be conducted.

Software Availability The iCanCloud simulator is Open Source (GNU General Public License version 3) and available at the following website: http://www.icancloudsim.org/

References 1. Foster, I., Zhao, Y., Raicu, I., Lu, S.: Cloud Computing and Grid Computing 360-Degree Compared. In: Proc. Grid Computing Environments Workshop, Austin, USA (2008) 2. Buyya, R., Yeo, C.S., Venugopal, S.: Market-oriented Cloud computing: Vision, hype, and reality for delivering IT services as computing utilities. In: Proc. of the Intl. Conference on High Performance Computing and Communications (HPCC), Dalian, China (2008) 3. Sterling, T.L., Stark, D.: A high-performance computing forecast: Partly cloudy. Computing in Science and Engineering 11, 42–49 (2009) 4. Vouk, M.A.: Cloud computing: Issues, research and implementations. In: Proc. of the 30th Intl. Conference on Information Technology Interfaces (ITI), Dubrovnic, Croatia (2008)

592

A. Nuñez et al.

5. Foster, I.T., Freeman, T., Keahey, K., Scheftner, D., Sotomayor, B., Zhang, X.: Virtual clusters for grid communities. In: Proc. of the Sixth Intl. Symposium on Cluster Computing and the Grid (CCGRID), Singapore (2006) 6. Sotomayor, B., Montero, R.S., Llorente, I.M., Foster, I.: Virtual Infrastructure Management in Private and Hybrid Clouds. IEEE Internet Computing 13, 14–22 (2009) 7. Nurmi, D., Wolski, R., Grzegorczyk, C., Obertelli, G., Soman, S., Youseff, L., Zagorodnov, D.: The eucalyptus open-source cloud-computing system. In: Proc. of the Sixth Intl. Symposium on Cluster Computing and the Grid (CCGRID), Shanghai, China (2009) 8. Buyya, R., Beloglazov, A., Abawajy, J.: Energy-efficient management of data center resources for cloud computing: A vision, architectural elements, and open challenges. In: Proc of the Intl. Conference on Parallel and Distributed Processing Techniques and Applications (PDPTA), Las Vegas, USA (2010) 9. Kim, K.H., Beloglazov, A., Buyya, R.: Power-aware provisioning of cloud resources for real-time services. In: Proc. of the 7th Intl. Workshop on Middleware for Grids, Clouds and e-Science, Urbana Champaign, Illinois, USA (2009) 10. The Network Simulator, NS-2, Web page http://www.isi.edu/nsnam/ns/ (date of last access: September 18, 2010) 11. Liu, J., Nicol, D.M.: DaSSF 3.1 User’s Manual, Dartmouth College (2001) 12. Varga, A.: The Omnet++ discrete event simulation system, In: Proc. of the European Simulation Multiconference (ESM), Prague, Czech Republic (2001) 13. OPNET modeller, Web page http://www.opnet.com/ (date of last access: September 18, 2010) 14. Miller, J.A., Nair, R.S., Zhang, Z., Zhao, H.: JSIM: A JAVA-based simulation and animation environment. In: Proc of the 30th Annual Simulation Symposium (ANSS), Atlanta, USA (1997) 15. Nún˜ ez, A., Fernández, J., Garcia, J.D., Garcia, F., Carretero, J.: New techniques for simulating high performance MPI applications on large storage networks. Journal of Supercomputing 51, 40–57 (2010) 16. Martin, M.M.K., Sorin, D.J., Beckmann, B.M., Marty, M.R., Xu, M., Alameldeen, A.R., Moore, K.E., Hill, M.D., Wood, D.A.: Multifacet’s general execution-driven multiprocessor simulator (GEMS) toolset. SIGARCH Computer Architecture News 33, 92–99 (2005) 17. Hardavellas, N., Somogyi, S., Wenisch, T.F., Wunderlich, R.E., Chen, S., Kim, J., Falsafi, B., Hoe, J.C., Nowatzyk, A.: Simflex: a fast, accurate, flexible full-system simulation framework for performance evaluation of server architecture. SIGMETRICS Performance Evaluation Review 31, 31–34 (2004) 18. Sulistio, A., Cibej, U., Venugopal, S., Robic, B., Buyya, R.: A toolkit for modelling and simulating Data Grids: An extension to GridSim. Concurrency and Computation: Practice and Experience 20, 1591–1609 (2008) 19. Bell, W.H., Cameron, D.G., Capozza, L., Millar, A.P., Stockinger, K., Zini, F.: Simulation of dynamic grid replication strategies in optorSim. In: Parashar, M. (ed.) GRID 2002. LNCS, vol. 2536, pp. 46–57. Springer, Heidelberg (2002) 20. Fujiwara, K., Casanova, H.: Speed and accuracy of network simulation in the simgrid framework. In: Proc. of the 1st Intl. Workshop on Network Simulation Tools (NSTools), Nantes, France (2007) 21. Liu, X.: Scalable Online Simulation for Modeling Grid Dynamics. PhD thesis, Univ. of California at San Diego (2004) 22. Sotomayor, B., Keahey, K., Foster, I.: Combining batch execution and leasing using virtual machines. In: Proceedings of the 17th International Symposium on High Performance Distributed Computing, HPDC 2008, pp. 87–96. ACM, New York (2008)


593

23. Calheiros, R.N., Ranjan, R., Beloglazov, A., De Rose, C.A.F., Buyya, R.: CloudSim: A toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms. Software: Practice and Experience (in press, accepted on June 14, 2010) 24. Buyya, R., Ranjan, R., Calheiros, R.N.: Modeling and Simulation of Scalable Cloud Computing Environments and the CloudSim Toolkit: Challenges and Opportunities. In: Proc. of the 7th High Performance Computing and Simulation Conference (HPCS), Kingston, Canada (2009) 25. Calheiros, R.N., Buyya, R., De Rose, C.A.F.: A heuristic for mapping virtual machines and links in emulation testbeds. In: Proc. of the Intl. Conference on Parallel Processing (ICPP), Vienna, Austria (2009) 26. Calheiros, R.N., Buyya, R., De Rose, C.A.F.: Building an automated and self-configurable emulation testbed for grid applications. Software: Practice and Experience 40, 405–429 (2010) 27. Schlesinger, S., et al.: Terminology for Model Creditibility. Simulation 32, 103–104 (1979) 28. Vazquez-Poletti, J.L., Barderas, G., Llorente, I.M., Romero, P.: A Model for Efficient Onboard Actualization of an Instrumental Cyclogram for the Mars MetNet Mission on a Public Cloud Infrastructure. In: Proc. of PARA: State of the Art in Scientific and Parallel Computing, Reykjavik, Iceland. LNCS (2010) (in press) 29. Harri, A., Linkin, V., Pichkadze, K., Schmidt, W., Pellinen, R., Lipatov, A., Vazquez, L., Guerrero, H., Uspensky, M., Polkko, J.: MMPM-Mars MetNet pre-cursor mission. In: European Geosciences Union General Assembly, Vienna, Austria (2008)