UMap: A fast, extensible memory-mapping library for diverse data storage ​

UMap is a unique, memory-centric architectural solution that solves key data movement challenges—making parallel computing faster, more cost-effective, and more energy efficient. ​

Unlike a standard system operation, which requires multiple steps to copy data from a data store into main memory, UMap creates a virtual memory space that provides direct access to all the data stores that a user needs to access, regardless of where the data physically resides. This setup makes it easier and faster for the processor to access the right data at the right time, and greatly reduces the burden on application developers.​

UMap is open source, free, and available to the public on GitHub.​ https://github.com/LLNL/umap

IM#: LLNL-VIDEO-864731