Data management in grid applications providers
Data management in grid applications providers
复制标题
网格应用程序提供商的数据管理
DOI：
10.1109/dfma.2005.19
复制
发表时间：
2005
期刊：
影响因子：
--
通讯作者：
L. Philippe
中科院分区：
文献类型：
--
作者：
Bruno Del;David Laiymani;J. Nicod;L. Philippe
应用服务提供者（ASP）模型是提供简单计算网格访问的一个有价值的候选模型。在此模型中，用户提交请求以在自己的数据上执行已部署的应用程序。在这种情况下，找到一个好的服务器来执行请求并不是提高性能的唯一方法。由于数据的大小可能很大，因此还必须避免在客户端和服务器之间进行无用的数据传输。现有的ASP平台大多提供数据管理服务，通过自动分发和传输数据来促进和改进执行。该文致力于研究两个问题.首先，它为ASP体系结构，特别是为DIET平台提供了一种有效的数据持久化策略。其次，它提出了一个与NetSolve数据管理模型类似的服务（而不是比较）。NetSolve方法提供了两种服务来管理平台内的数据：分布式存储基础架构，用于提供数据传输和存储;以及请求排序，用于减少客户端和服务器之间的网络流量。DIET数据管理模型基于两个关键点：数据标识符和数据树管理器（DTM）。为了避免从客户端到服务器的相同数据的多次传输，DTM允许在计算之后将数据留在平台内，而数据标识符由客户端进一步使用以引用其数据。在后面的部分中，我们将展示这些不同的方法如何在客户端产生不同的编程模型和执行性能。
The application service provider (ASP) model is a valuable candidate to provide simple computing grid access. In this model, users submit requests to execute already deployed applications on their own data. In this context, finding a good server to execute a request is not the only way to increase performances. As the size of the data may be huge, it is also mandatory to avoid useless data transfers between client and servers. Most of existing ASP platforms provide data management services to facilitate and to improve executions by automatically distributing and transferring data. The aim of this paper is twofold. First, it presents an efficient data persistence policy for ASP architectures and particularly for the DIET platform. Second, it proposes a parallel (not a comparison) of this service with the NetSolve data management model. The NetSolve approach provides two services to manage data inside the platform: the distributed storage infrastructure, used to provide data transfer and storage and the request sequencing used to decrease network traffic amongst client and servers. The DIET data management model is based on two key points: the data identifiers and the data tree manager (DTM). To avoid multiple transmissions of the same data from a client to a server, the DTM allows to leave data inside the platform after computations while data identifiers are used further by the client to reference its data. In the remainder, we show how these different approaches lead to different programming models and execution performances on the client side.