Accelerated Networking with H-series VMs on Azure for older OS distributions
The accelerated networking update for the HPC SKUs on Azure has caused problems for older OS distributions or any MPI versions that do not use the latest UCX. This is due to inconsistent naming for the IB devices. My recent patch to rdma-core can be used to provide consistent naming with udev rules. This following script can be used when building an image:
yum install -y cmake libnl3-devel git clone https://github.com/linux-rdma/rdma-core.git cd rdma-core bash build.sh cp build/bin/rdma_rename /usr/lib/udev/ cat <<EOF >/etc/udev/rules.d/60-ib.rules # Accelnet board ACTION=="add", ATTR{board_id}=="MSF0010110035", SUBSYSTEM=="infiniband", PROGRAM="rdma_rename %k NAME_FIXED mlx5_an0" # HBv2 board ACTION=="add", ATTR{board_id}=="MT_0000000223", SUBSYSTEM=="infiniband", PROGRAM="rdma_rename %k NAME_FIXED mlx5_ib0" # HC board ACTION=="add", ATTR{board_id}=="MT_0000000010", SUBSYSTEM=="infiniband", PROGRAM="rdma_rename %k NAME_FIXED mlx5_ib0" EOF
This will name the accelerated networking mlx5_an0 and the infiniband to mlx5_ib0. Now, you can use the older MPI/UCX versions by setting:
export UCX_NET_DEVICES=mlx5_ib0:1
The script includes rules that will work for HB, HC, HBv2 and NDv2.
Published on:
Learn moreWe can help you with Accelerated Networking with H-series VMs on Azure for older OS distributions
If you want help implementing, troubleshooting, or improving this product, contact us and we’ll point you in the right direction.
Related posts
Large-scale docking for drug design on Azure
“Only 10 total drugs in 46 years have been intentionally developed for childhood cancer and have reached FDA approval (reference). Childhood ...
E2E deployment of a production ready NDv4 (A100) cluster targeting large deep learning training
Introduction The NDv4 series is very popular for running large deep learning training jobs, which require lots of floating-point perfo...
Performance considerations for large scale deep learning training on Azure NDv4 (A100) series
Background The field of Artificial Intelligence is being applied to more and more application areas, such as self-driving cars, natural langua...
Best Practice for Running Cadence Spectre X on Microsoft Azure
Co-authors: Richard Paw and Andy Chan Electronic Design Automation (EDA) consists of a set of software (tools) and workflows for design...
Spack in a Multi-User HPC Environment on Azure
Spack is a package management tool designed for HPC environments. In this article we will demonstrate how to install and configure Spack...
AzureHPC Lustre Marketplace Offer
The AzureHPC Lustre Marketplace offer provides a fast and easy way to deploy a Lustre Filesystem on Azure. The features include: Deplo...
Mainframe Migration to Microsoft Azure with Intel Architecture on Your Terms.
IntroductionMainframes are still in wide use today for business-critical applications in many industries. According to a 2021 survey by IBM, 7...
Why Organizations are Entrusting Microsoft Cloud with their Mainframes
There is a resurgence of interest and renewed priority to transform core enterprise systems. This resurgence is largely driven by rapidly shif...
Automated HPC/AI compute node health-checks Integrated with the SLURM scheduler
Overview It is best practice to run health-checks on compute nodes before running jobs, this is especially important for tightly coupl...
GPU Monitoring using Azure Monitor
Overview Today, many highly parallel HPC/AI applications use GPU to improve the run-time performance. It is important to be able to monitor t...