Unisys Corporation

Robert Len Walker

Matthew Miller

An embodiment of the invention is a technique to affinitize a thread in a multiprocessor system having N clusters of processors. A usage status of an active thread in the multiprocessor system is determined. The active thread utilize one of first and second local memories associated with first and second processor configurations, respectively. One of the first and second processor configurations is selected based on the usage status.

Selecting processor configuration based on thread usage in a multiprocessor system

An embodiment of the invention is a technique to affinitize a thread in a multiprocessor system having N clusters of processors. A count threshold is obtained for thread affinity of the thread. A first thread count is determined. The thread is affinitized to a first cluster in the N clusters of processors according to the first thread count and the count threshold.

Affinitizing threads in a multiprocessor system

An embodiment of the invention is a technique to generate a mask schedule for thread scheduling in a multiprocessor system having N clusters of processors. A primary mask is obtained from a primary set of granularity masks. The granularity masks are created according to a thread granularity. The primary mask is filtered using a filter mask to generate a first affinity mask.

Dynamically generating masks for thread scheduling in a multiprocessor system

A method and a system is disclosed for managing distribution of computer-executable program threads between a plurality of central processing units (CPUs) administered by an operating system in a multi-CPU computer system having a plurality of memory caches shared amongst the CPUs. The method includes assigning the CPUs to a plurality of CPU-groups of a predetermined group-size, selecting a CPU-group form the CPU-groups, setting a predetermined threshold for the selected CPU-group, and affinitizing a program thread to the selected CPU-group based on the predetermined threshold wherein the operating system distributes the program threads among the CPU-groups based on the affinitizing. In this way, the memory access time delays associated with the transfer of data amongst the CPU-groups can be advantageously reduced while the ability to more effectively utilize the total number of available CPUs in the processing of the program threads is advantageously increased.

Method and system for managing distribution of computer-executable program threads between central processing units in a multi-central processing unit computer system

Growing community of inventors

Mission Viejo, CA, United States of America

Robert Len Walker