Skip to Main Content
Modern software systems are complex and increasingly vulnerable to malicious attack. In order to apply bug fixes and protect against security weaknesses, workstation administrators must continuously patch operating system and application program installations. To simplify this process, administrators generally configure systems into clusters with common installations of operating systems and application programs in a manner that patches and other routine maintenance can be applied en masse. Two widely used approaches offer complementary advantages: Installations and updates are particularly convenient when all data including operating system and programs is served by traditional centralized network-connected file servers (networked file systems) because administrators need only maintain a single image that is distributed online to all workstations. In contrast, operating systems and application programs can be installed onto disk drives within each workstation. This mass replication of common data provides high aggregate bandwidth through parallelism since each disk operates independently. However, in contrast to the network file system approach, it is substantially more difficult to keep a large network of systems up to date when each system has an autonomous installation. Through the aggressive use of cooperative file caching, we expect that sloth-NFS is will provide the advantages of both approaches. In this paper, we discuss the administrative problems with current distribution methods for program and operating system installations, analyze design decisions necessary in Sloth-NFS and present results from an initial simulator experiment.