دانلود رایگان ترجمه مقاله توزیع کد برنامه دینامیک در ابرهای فراساختار به شکل یک خدمات (نشریه IEEE 2013)

توزیع کد برنامه دینامیک در ابرهای فراساختار به شکل یک خدمات

Dynamic Program Code Distribution in Infrastructure-as-a-Service Clouds

Abstract

Elastically scaling cloud computing applications are becoming more and more prevalent in today’s IT landscapes. One problem of building such applications in an Infrastructureas-a-Service cloud is the runtime distribution of program code, configuration files and other resources. While it is possible to include all required program code in the used IaaS base images, this severely restricts the achievable dynamicity at runtime. In this paper, we present a framework for dynamic program code distribution. Our approach handles code distribution entirely transparently on middleware layer. We base our solution on an existing middleware, CloudScale. The paper discusses the design and implementation of our code distribution approach on top of CloudScale, and numerically evaluates the practicability and performance of the approach based on an illustrative case study.

1 Introduction

In the last years, the advancement of cloud computing [16] has transformed the entire IT industry, and has given new opportunities and abilities to developers and users. Moreover, cloud computing simplifies the implementation of innovative ideas for small companies or individuals, and lowers production and maintenance costs for industrial applications [1]. Because of cloud computing, elasticity of resources becomes a feature of applications instead of a description of data centers. The notion of elasticity for cloud applications morphs into elasticity of cost, resource and quality [10], giving additional dimensions and abilities for developers to optimize application and service provisioning.

Cloud computing is a promising technological choice for new application development projects, but if an application already exists, migration costs have to be considered before the advantages of the cloud can be leveraged. Cloud migration is known to be a challenging problem [12]. Often, migration reveals architectural problems and requires refactoring or redesigning of applications. However, even when the application architecture already fits the cloud computing paradigm, some amount of additional work is required in order to let the application fully benefit from the cloud.

The most basic cloud service model is the Infrastructureas-a-Service (IaaS) approach [16]. On this level, cloud service providers offer virtual machines with requested configuration and operation system (usually in a form of hard drive image) to satisfy application computation requirements [5]. This layer is preferable for cloud migration as it requires less migration effort (and is better standardized) than Platformas-a-Service [16] (PaaS). When an IaaS application has to scale up (i.e., use more virtual machines than before), one problem is how the availability of the current version of the application code, configuration files and other resources can be ensured on the new host. In the following, we will use the term “program code” as shorthand for the application code and all dependent files. The trivial approach is to include this program code in the virtual machine base image, but this approach is reasonable only in situations when it is entirely static and will not be modified during application lifetime. However, reallife applications are typically not quite as static. Instead, the program code often evolves over time, and multiple different versions of an application have to be executable in parallel. In such scenarios, hardcoding the program code and other files into the virtual machine images becomes complicated or even impossible. An alternative way to achieve program code distribution is to include facilities for dynamic code search and distribution on middleware level.

This paper introduces a framework for seamless runtime program code distribution. The framework is based on our earlier CloudScale [15] research prototype, but due to high independence level, can be used separately in other systems as well. CloudScale is a middleware, which simplifies the development of Java application in an IaaS cloud. In our framework, program code distribution is entirely handled by the underlying CloudScale middleware. We evaluate different configuration and implementation options, and present numerical performance results based on an illustrative case study.

The remainder of this paper is structured as follows. Section II describes illustrative case study that demonstrates the code distribution problem based on a real-world application. Section III describes research related to our work. Afterwards, Section IV presents the background of the approach we present in this article, most importantly, the CloudScale framework. Section V presents the actual contribution of this paper, which is consequently evaluated in Section VI. Finally, the paper is concluded in the Section VII.