Gerenciamento de Recursos Distribuídos em Sistemas de Grande Escala

Jeferson Roberto MarquesFabio Kon

With the popularization of computer networks in the past decade, local area networks composed of dozens of powerful machines became a commonplace in academic, commercial, and governmental institutions. These networks are often connected to the Internet and offer a wide variety of services to its users. Nevertheless, despite the almost two decades of research in distributed processing and load balancing, these systems are almost always used in a simplistic and inefficient way. A few services (such as file systems and the web) are distributed according to a conventional client/server model where the servers are almost always fixed. Most applications (such as text editors, calendars, spreadsheets, web browsers) are almost always executed locally in the client's machine even if it is not the most appropriate location to execute it and if there are other more powerful machines that are idle. A consequence of this behavior is that the resources in the distributed system are often underutilized and the quality of service offered to the user is not satisfactory. The work presented in this paper revisits the concept of load balancing of user applications, presenting two new contributions. First, we address the problem with a novel architecture based on distributed objects and the CORBA standard. Second, we describe new protocols that are scalable to wide-area networks and can be used tointegrate dozens or hundreds of heterogeneous local area networks,supporting the provision of quality of service. Thus, when logging into a machine connected to the network, the user gets access not only to the local machine but to the whole distributed system in all its power and diversity of resources.

Caso o link acima esteja inválido, faça uma busca pelo texto completo na Web: Buscar na Web

Biblioteca Digital Brasileira de Computação - Contato:
     Mantida por: