DragonHPC
DragonHPC is a programmable distributed runtime for HPC & AI workflows. The DragonHPC distributed runtime will power complex post-exascale workflows that can scale, that are cloud-native, that can access data efficiently, that can span multiple systems, and that can be used with multiple languages on heterogeneous hardware.
Key advantages of using DragonHPC
Among other frameworks and libraries, DragonHPC addresses much of the friction developers and scientists face when approaching a computational problem:
Program across widely distributed resources (e.g., laptops, servers, cloud, supercomputers) as if they were all one computer.
Implement applications and workflows across Python, C, C++, and Fortran with interoperable high-performance objects instead of being locked into a single language.
Implement Python-based applications and workflows using the standard multiprocessing API instead of non-standard interfaces.
Develop applications and highly dynamic workflows without the limitations of static execution graphs.
Orchestrate individual or collections of Python functions, binaries, and MPI processes.
Leverage RDMA and leading-edge HPC communication techniques through simple high-level interfaces and communication objects instead of communication through a filesystem.
Users can also access DragonHPC at any level of its architecture. This allows users to develop new components within Dragon that natively interoperate with existing components. Flexibility, composability, and the ability to adopt DragonHPC for applications, tools, and workflows with strict performance requirements dramatically improves user productivity.
Online resources for getting up and running with DragonHPC
Related Blogs
