Content area
Full Text
We propose a name service that enables construction of a uniform, global, hierarchical namespace, a key feature needed to create a file-system grid. Combined with other grid replication and location-lookup mechanisms, it supports independence of position for users and applications as well as transparency of data location in a scalable and secure fashion. This name service enables federation of individual files as well as file-system trees that are exported by a variety of distributed file systems and is extensible to include non-file-system data such as databases or live data feeds. Such a federated namespace for files can be rendered by network file servers, such as NFS (Network File System) or CIFS (Common Internet File System) servers, proxies supporting the NAS (network-attached storage) protocol, or grid data service interfaces. File access proxies, which handle protocol translation, can also include caching and replication support to enhance data access performance. A uniform namespace with global scope and hierarchical ownership allows sharing file data between and within organizations without compromising security or autonomy.
The major goal of grid computing is to foster sharing of widely distributed resources. All distributed systems are faced with network limitations and scaling. The grid shares these fundamental problems with other distributed systems, but unlike more tightly coupled systems, such as clusters, these problems cannot be eliminated, but must be accepted as part of the environment in which the grid operates. Crucial network problems include large latencies and low bandwidths; the question is not how to improve these parameters as much as how can the system accommodate these limitations. The large scale of grid systems expands the boundaries containing current systems in the number of applications, users, groups, domains, and hosts, the variety of their operating systems, and the diversity of protocols. A key result of this scale is that the security mechanisms of small groups break down, and stronger and more formal measures are needed. Security is also important because owners are more willing to share their resources when they can maintain control over them.
Files represent an especially fundamental resource for collaboration within grid virtual organizations.1 The usage of file data is strongly impacted by features of the distributed environment, especially latency, bandwidth, and hot spots, to name a few. The...