MVAPICH2 on Azure HPC Clusters

This post has been republished via RSS; it originally appeared at: Azure Compute articles.

The MVAPICH2/MVAPICH2-X MPI libraries offer high performance, scalability, and fault tolerance for high-end computing systems and servers using the InfiniBand network. These libraries help HPC users to extract the potential of networking technologies for modern systems and are powering several supercomputers in the TOP500 list.

 

Azure HPC team is happy to announce the availability of optimized MVAPICH2/MVAPICH2-X MPI libraries in Azure HPC VM images. Customers can deploy any of the Azure HPC images and can drive the best performance using MVAPICH libraries.

 

Feature Highlights - MVAPICH2-Azure-2.3.3:

    - Based on MVAPICH2-2.3.3

    - Enhanced tuning for point-to-point and collective operations

    - Targeted for Azure HB, HBv2, and HC virtual machine instances

 

Feature Highlights - MVAPICH2-X-Azure-2.3.3:

    - Based on MVAPICH2-X-2.3rc3

    - Support for advanced MPI features (Direct Connect, Cooperative Protocol) and XPMEM

    - Targeted for Azure HB, HBv2, and HC virtual machine instances

 

Performance Expectations:

The following figures highlight the MPI latency and bandwidth results using MVAPICH2 and MVAPICH2-X. These were taken on Azure HBv2 VM instances running Azure CentOS 8.1 HPC image.

 

osu_latency-small-mv2.PNG

 

 

 

osu_bw-mv2.PNG

 

Additional performance results are available from here: http://mvapich.cse.ohio-state.edu/performance/mv2-azure-pt_to_pt/.

 

Deploy MVAPICH2 on Azure:

As described above, MVAPICH2/MVAPICH2-X is available in Azure HPC VM images. These images are available from Azure Marketplace, and it can be deployed through a variety of deployment vehicles (CycleCloud, Batch, ARM templates, etc).  AzureHPC scripts provide an easy way to quickly deploy an HPC cluster using these HPC VM images.

 

Leave a Reply

Your email address will not be published. Required fields are marked *

*

This site uses Akismet to reduce spam. Learn how your comment data is processed.