py-libnuma 1.2

Python3 libnuma ctypes wrapper

py-libnuma

py-libnuma is python3 interface to numa Linux library so that you can set task affinity and memory affinity in python level for your process which can help you to improve your code's performence.

Installation

pip install py-libnuma

Usage

py-libnuma categorize libnuma's apis into 3 groups :memory, schedule and info. You can set your tasks' cpu affinity, memory affinity and get information about your systems's hardware with these 3 modules respectively. For more information about APIs, you can refer to API.md in github

schedule

numa.schedule helps you to set cpu affinity for your process, if you have multiple numa nodes, and you want your process to be scheduled on cpus from node1 and node2, you can use numa.schedule like this

from numa import schedule
schedule.run_on_nodes(1,2)

If you want a certain process with pid to run on specific cpus, you can use numa.schedule like this

from numa import schedule
schedule.run_on_cpus(pid, 1,3,4,6)

memory

numa.memory helps you to set memory policy for your process, if you want your current process to allocate memory from multiple numa nodes to balance local and remote memory access, you can use numa.memory like this:

from numa import memory    
memory.set_interleave_nodes(0,1)

or you can make your process to allocate from certain nodes by

from numa import memory    
memory.set_membind_nodes(1)

Info

numa.info helps you to get information about your numa hardware, you can check hardware information by:

from numa import info    
info.numa_hardware_info()

This will tell you distance between different numa nodes and node-cpu relation. You can also check cpu set for certain nodes by:

from numa import info    
info.node_to_cpus(1)

or check which node is a certain cpu belongs to by:

from numa import info    
info.cpu_to_node(1)

For more information about APIs, you can refer to API.md in github

Feedback

If you have any problems with using this package, feel free to create issues and you will get answered in no more than 24 hours