Achieving Large Scale Parallelism Through Operating System Resource Management on the Intel TFLOPS Supercomputer Export

@misc{citeulike:6060462, abstract = {From the point of view of an operating system, a computer is managed and optimized in terms of the application programming model and the management of system resources. For the TFLOPS system, the problem is to manage and optimize large scale parallelism. This paper looks at the management in terms of three key topics: memory management, communication, and input/output. For memory management, we discuss some of the design decisions made including the appropriate use of demand paged virtual memory in the system. For communication, we describe the software protocols and interactions that permit a system of 4500 nodes to approach the maximum hardware performance. For I/O, we look at the problem of funneling data from many computation nodes to a small number of external devices.}, author = {Garg, Sharad and Godley, Robert and Griffiths, Richard and On The Intel Tflops Supercomputer}, citeulike-article-id = {6060462}, citeulike-linkout-0 = {http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.19.5446}, keywords = {opsystem, parallel, resources}, posted-at = {2009-11-03 15:13:44}, priority = {3}, title = {Achieving Large Scale Parallelism Through Operating System Resource Management on the Intel TFLOPS Supercomputer}, url = {http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.19.5446}, year = {1998} }

TY - ELEC ID - citeulike:6060462 L3 - citeulike-article-id:6060462 N2 - From the point of view of an operating system, a computer is managed and optimized in terms of the application programming model and the management of system resources. For the TFLOPS system, the problem is to manage and optimize large scale parallelism. This paper looks at the management in terms of three key topics: memory management, communication, and input/output. For memory management, we discuss some of the design decisions made including the appropriate use of demand paged virtual memory in the system. For communication, we describe the software protocols and interactions that permit a system of 4500 nodes to approach the maximum hardware performance. For I/O, we look at the problem of funneling data from many computation nodes to a small number of external devices. TI - Achieving Large Scale Parallelism Through Operating System Resource Management on the Intel TFLOPS Supercomputer KW - opsystem KW - parallel KW - resources AU - Garg, Sharad AU - Godley, Robert AU - Griffiths, Richard AU - On The Intel Tflops Supercomputer PY - 1998/// UR - http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.19.5446 ER -