Posted On: Jul 19, 2019
The Elastic Fabric Adapter (EFA) provider is officially integrated into the Libfabric 1.8 release. Customers can directly use Libfabric 1.8 without needing to install the EFA provider separately.
EFA is a network interface for Amazon EC2 instances that enables customers to run applications requiring high levels of inter-node communications at scale on AWS. With EFA, High Performance Computing (HPC) applications using the Message Passing Interface (MPI) and Machine Learning (ML) applications using NVIDIA Collective Communications Library (NCCL) can scale to thousands of CPUs or GPUs.
The Libfabric library is the preferred interface for accessing the EFA hardware. Libfabric is a core component of OpenFabrics Interfaces (OFI) and the library that defines and exports the user-space API of OFI. It is typically the only software that applications deal with directly.
To learn more about EFA, please visit EFA documentation. To launch a HPC cluster with EFA, check out here.