International Business Machines Corporation

Dharmesh J Patel

Daniel Paul Kolz

Jay Symmes Bryant

Nicholas Bruce Goracke

A method and apparatus improves the block allocation time in a parallel computer system. A pre-load controller pre-loads blocks of hardware in a supercomputer cluster in anticipation of demand from a user application. In the preferred embodiments the pre-load controller determines when to pre-load the compute nodes and the block size to allocate the nodes based on pre-set parameters and previous use of the computer system. Further, in preferred embodiments each block of compute nodes in the parallel computer system has a stored hardware status to indicate whether the block is being pre-loaded, or already has been pre-loaded. In preferred embodiments, the hardware status is stored in a database connected to the computer's control system. In other embodiments, the compute nodes are remote computers in a distributed computer system.

Block allocation times in a computer system

Embodiments of the present invention provide a method for reducing load time of a program in a highly-parallelized or distributed computer. In one embodiment, this is accomplished by selectively reusing entries in a page table generated during a previous invocation of the program at a particular compute node of the highly-parallelized or distributed computer system.

Process restart on a compute node

Embodiments of the invention provides techniques for defragmenting blocks of resources allocated to perform computing jobs on a distributed or clustered system so that more contiguous physical resources may be made available to users submitting new job requests. Typically, the defragmentation process is performed when a job is submitted that requires access to a computing block that is larger than any currently available block in the parallel computing system.

Defragmenting blocks in a clustered or distributed computing system

A memory management mechanism a nodal having multiple processors in a massively parallel computer system dynamically configures nodal memory on demand. A respective variable-sized subdivision of nodal memory is associated with each processor in the node. A processor may request additional memory, and the other processor(s) may grant or veto the request. If granted, the requested memory is added to the subdivision of the requesting processor. A processor can only access memory within its own subdivision. Preferably, each subdivision contains a daemon which monitors memory usage and generates requests for additional memory.

Memory request/grant daemons in virtual nodes for moving subdivided local memory space from VN to VN in nodes of a massively parallel computer system

Memory request / grant daemons in virtual nodes for moving subdivided local memory space from VN to VN in nodes of a massively parallel computer system

Growing community of inventors

Rochester, MN, United States of America

Dharmesh J Patel