Abstract: A parallel computer system comprising a node group having numbers of nodes connected by a network, in which a job scheduler of a job server that schedules jobs to be executed by a node of the node group comprises a temperature calculating unit which with a node being used of the node group as an imaginary heat source and with the assumption that a quantity of heat is conducted from the heat source to a surrounding node, calculates a temperature of a surrounding free node based on a distance from the heat source, a free region extracting unit which selects, from a plurality of temperature groups obtained by grouping free nodes on a certain temperature range basis, a temperature group meeting the number of free nodes required by a job according to a temperature and takes out a lowest temperature free node from the selected temperature group as a center node, and a node selecting unit which sequentially selects the necessary number of free nodes starting with a shortest distance free node centered around the cen

Abstract: System and method to perform rapid I/O processing for a large file reducing load on a server computer and a network. The client computer accesses a disk device directly to perform input/output processing relating to a file operation through a channel switch based on real file information obtained by the client computer from a parent server computer upon a file-open operation. Further, a child server computer manages the disk device, which enables different real files in a same virtual file to be processed an I/O at the same time by a plurality of client computers.

Abstract: A computer system is provided with a file system storing data; a plurality of I/O nodes which are adapted to access the file system; a compute node adapted to execute a job and to issue an I/O request when requiring an I/O operation; and a job server for job scheduling which dynamically allocates an I/O resource of the I/O nodes to a job without stopping execution of the job. The job server includes an I/O node scheduler adapted to, when being not able to fully secure an desired amount of the I/O resource of the I/O nodes required by the job in starting the job, secure a part of the required amount of the I/O resource of the I/O nodes, and to allocate the secured part of the I/O resource to the job.

Abstract: A parallel computer system comprising a node group having numbers of nodes connected by a network, in which a job scheduler of a job server that schedules jobs to be executed by a node of the node group comprises a temperature calculating unit which with a node being used of the node group as an imaginary heat source and with the assumption that a quantity of heat is conducted from the heat source to a surrounding node, calculates a temperature of a surrounding free node based on a distance from the heat source, a free region extracting unit which selects, from a plurality of temperature groups obtained by grouping free nodes on a certain temperature range basis, a temperature group meeting the number of free nodes required by a job according to a temperature and takes out a lowest temperature free node from the selected temperature group as a center node, and a node selecting unit which sequentially selects the necessary number of free nodes starting with a shortest distance free node centered around the cen

Abstract: A computer system is provided with a file system storing data; a plurality of I/O nodes which are adapted to access the file system; a compute node adapted to execute a job and to issue an I/O request when requiring an I/O operation; and a job server for job scheduling which dynamically allocates an I/O resource of the I/O nodes to a job without stopping execution of the job. The job server includes an I/O node scheduler adapted to, when being not able to fully secure an desired amount of the I/O resource of the I/O nodes required by the job in starting the job, secure a part of the required amount of the I/O resource of the I/O nodes, and to allocate the secured part of the I/O resource to the job.

Abstract: To perform rapid I/O processing for a large file reducing load on a server computer and a network. The client computer accesses a disk device directly to perform input/output processing relating to a file operation through a channel switch based on real file information obtained by the client computer from a parent server computer upon a file-open operation. Further, a child server computer manages the disk device, which enables different real files in a same virtual file to be processed an I/O at the same time by a plurality of client computers.

Abstract: A file sharing system comprises a client for transmitting only control information on I/O to a server and, based on I/O target determination information received from the server, performing I/O by itself with a disk device, a server for receiving control information from the client and transmitting to the client information for determining the disk device targeted for the I/O or information for notifying the performance of I/O termination processing, and a disk device connected to the client and the server through a network line, comprising a protocol conversion means for taking out a command based on an I/O protocol such as SCSI from the I/O information converted into a format suitable for the I/O protocol and received from the client.

Abstract: A process stop method and apparatus applicable to a distributed memory multi-processor system allows an efficient stop processing for the entry of a checkpoint during parallel processing where data are communicated between different nodes. The system comprises a plurality of nodes interconnected through a network, each of which is provided with a thread for parallel processing. Each of the nodes has a management process to manage its own thread engaged in parallel processing. When a request for a checkpoint is dispatched as an external command, the management process of a node whose node number is the smallest requests a stop from the thread of the same node. After confirming that the thread in question has stopped, the management process delivers a stop request to the management process of a node whose node number is the next smallest.