| Dynamic self-organized architecture based on grid component (DSAG) is a new architecture of HPC. It divides the different functional components of a traditional computer apart and forms all kinds of grid-components, which provide service independently via optical interconnection network. Then the HPC system will be built with grid-components dynamically according to the application requirements. In DSAG, the computer is a global system. Accordingly, the traditional operating system functions should be reconsidered from the global scope.This paper presents the reseach on operating system with light-weight kernel under DSAG environment, which can support science computing applications effectively. It is made up by Sand OS and Remote Server. Sand OS is a light-weight kernel which provides process management and memory management for computing nodes of DSAG. Remote Server is a file server which provides file system management. The structure of this operating system shows the idea of operating system function distribution. Remote system call mechanism is presented in this paper, which can reorganize the loose coupled operating system functions and then provide some complete standard API to support representative applications. In addition, this mechanism can also be applied to the construction of sigal system image.Sand OS is discussed in this paper, whose design goals are simplified, effective and multi-architecture support. We adopted some reasonable ways to simplify this kernel, and then reduced the influence on program execution.Then the design of remote system call mechanism is discussed under DSAG environment, and the concept of virtual file descriptor is presented, which could share the file descriptors of Remote Server among multiple client processes effectively. Using this mechanism, some kinds of sigle system image functions were implemented.Finally, this paper discusses the construction of a Linux application environment with high compatibility for Sand OS. This environment supports the traditional serial programming mode and some important parallel ones, e.g. OpenMP/pthread and MPI. It also supports many representative programs. |