NumPy Persistence to Non-Volatile Memory

This library will help you persist numpy array structures into NVM devices.

Internally, it uses pynvm and buffers --which are natively supported by numpy.

Installation

pip install npp2nvm

Requirements

• A NVM device, such as Intel(R) Optane(TM) DC Persistent Memory modules.
• Python 3.x
• numpy arrays to be stored.

Usage

Two environment variables are used for configuring the storage:

• NPP2NVM_SIZE which specifies the size (in MiB) that will be assigned to persistence
• NPP2NVM_PATH which specifies the file path that will be used for persistence. Note that this file should be located into a Storage Class Memory device, which will allow pynvm to leverage its characteristics --keep in mind that pynvm are bindings of the PMDK libraries.

Example

import numpy as np
import npp2nvm

a = np.random.random([256, 256])
a = npp2nvm.np_persist(a)
print(a.shape)
# > (256, 256)

Limitations

Storage and retrieval of data structures (outside the application lifecycle) are not implemented in this library (yet?). You can implement your own mechanism and/or propose a PR. It will be appreciated!