Affinity-aware parallel zeroing of memory in non-uniform memory access (NUMA) servers

Patent metadata

Field	Value
Publication number	US-9983642-B2
Application number	US-201614987151-A
Country	US
Kind code	B2
Filing date	Jan 4, 2016
Priority date	Jun 25, 2015
Publication date	May 29, 2018
Grant date	May 29, 2018

Abstract

A method for zeroing memory in computing systems where access to memory is non-uniform includes receiving, via a system call, a request to delete a memory region. The method also includes sorting, after receiving the request, one or more pages of the memory region according to each associated affinity domain of each page. The method further includes sending requests to zero the sorted one or more pages to one or more zeroing threads that are attached to the respective affinity domain. The method further yet includes waiting, after sending the requests, to return to the system caller until a message is received, from the zeroing threads in each affinity domain, indicating that all the page zeroing requests have been processed.

First claim

What is claimed is:

1. A method for zeroing pages of memory in a computing system where access to memory is non-uniform, the method comprising:
receiving, via a system call, a request to delete a memory region;
sorting, after receiving the request, one or more pages of the memory region according to each associated affinity domain of each page, wherein each affinity domain comprises a cluster of processors and memory local to the cluster of processors;
sending requests to zero the sorted one or more pages to one or more software threads that are attached to the respective affinity domain; and
waiting, after sending the requests, to return to a system caller until a message is received, from the software threads in each affinity domain, indicating that all requests to zero have been processed.

2. The method of claim 1 , wherein, before receiving the request to delete the memory region, the method further comprises:
analyzing a distribution of one or more processors and memory in the computing system; and
identifying each affinity domain, based on the analysis.

3. The method of claim 1 , wherein the one or more software threads zero the sorted one or more pages in parallel.

4. The method of claim 1 , wherein sending the requests to zero the sorted one or more pages to the one or more software threads comprises sending the requests to zero to request queues in each affinity domain.

5. A method for zeroing pages of memory in a computing system where access to memory is non-uniform, the method comprising:
receiving, via a system call, a request to delete a memory region;
sorting, after receiving the request, one or more pages of the memory region according to each associated affinity domain of each page, wherein each affinity domain comprises a cluster of processors and memory local to the cluster of processors, and wherein the local memory in each affinity domain has the same access characteristics for each of the processors in the cluster;
sending requests to zero the sorted one or more pages to one or more software threads that are attached to the respective affinity domain; and
waiting, after sending the requests, to return to a system caller until a message is received, from the software threads in each affinity domain, indicating that all requests to zero have been processed.

6. A method for zeroing pages of memory in a computing system where access to memory is non-uniform, the method comprising:
analyzing a distribution of one or more processors and memory in the computing system, wherein analyzing the distribution of the one or more processors and the memory in the computing system occurs upon at least one of the following: an addition of physical memory, removal of physical memory, addition of a processor, or removal of a processor;
identifying, based on the analysis, one or more affinity domains, wherein each affinity domain comprises a cluster of processors and memory local to the cluster of processors;
receiving, via a system call, a request to delete a memory region;
sorting, after receiving the request, one or more pages of the memory region according to each associated affinity domain of each page;
sending requests to zero the sorted one or more pages to one or more software threads that are attached to the respective affinity domain; and
waiting, after sending the requests, to return to a system caller until a message is received, from the software threads in each affinity domain, indicating that all requests to zero have been processed.

Assignees

• IBM

Inventors

• Hedge Nikhil
• Mealey Bruce
• Rogers Mark D

Classifications

• G06F3/0652

  Erasing, e.g. deleting, data cleaning, moving of data to a wastebasket · CPC title

• G06F3/0644

  Management of space entities, e.g. partitions, extents, pools · CPC title

• G06F3/0608

  Saving storage space on storage systems · CPC title

• G06F2009/45583

  Memory management, e.g. access or allocation · CPC title

• G06F3/0673

  Single storage device · CPC title