Migrating threads across NUMA nodes using corresponding page tables and based on remote page access frequency

摘要

Implementations provide for tracking memory accesses to optimize processor task placement is disclosed. A method includes creating a page table (PT) hierarchy associated with a thread, wherein the PT hierarchy comprises identifying information of memory pages and access bits corresponding to each of the memory pages, setting the respective access bit of one or more of the memory pages accessed by the thread while the thread is executing, collecting access bit information from the PT hierarchy associated with the thread, wherein the access bit information comprises the set access bits in the PT hierarchy, determining, in view of the collected access bit information, memory access statistics for the thread, and utilizing, during runtime of the thread, the memory access statistics for the thread in a determination of whether to migrate the thread to another processing device during the runtime of the thread.

主权项

1. A method, comprising:
creating, by a processing device, a page table (PT) hierarchy associated with a thread to be run on the processing device, wherein the PT hierarchy comprises:
information identifying each memory page maintained in addressable memory space accessible by the thread; andaccess bits corresponding to each of the memory pages; utilizing, by the processing device, the PT hierarchy associated with the thread to identify locations of memory pages to access during execution of the thread by the processing device; setting the respective access bits of the memory pages accessed by the thread while the thread is executing on the processing device; collecting access bit information from the PT hierarchy associated with the thread, wherein the access bit information comprises the set access bits in the PT hierarchy; determining, in view of the collected access bit information, memory access statistics for the thread; and utilizing, by the processing device during runtime of the thread, the memory access statistics for the thread in a determination of whether to migrate the thread to another processing device during the runtime of the thread.