在批处理作业中利用带有OpenMPI的RDMA网络 [英] Leveraging RDMA network with OpenMPI in Batch jobs
问题描述
你好
在Technet入门文章(https://blogs.technet.microsoft.com/windowshpc/2016/07/20/introducing-mpi-support-for-linux-on-azure-batch/)中，作者声称
In the introductory technet article (https://blogs.technet.microsoft.com/windowshpc/2016/07/20/introducing-mpi-support-for-linux-on-azure-batch/) the authors claim that
通过创建A8或A9计算节点池，批处理MPI任务可以充分利用那些Azure VM的高速，低延迟RDMA网络."
"By creating a pool of A8 or A9 compute nodes, Batch MPI tasks can fully leverage the high-speed, low-latency RDMA network for those Azure VMs."
它还指出池中的所有节点都应安装MPI(OpenMPI，IntelMPI或任何其他MPI安装程序)".据此我了解到，只要您有一个A8/A9节点池，并且您使用的是已安装任何MPI的映像， 然后自动节点将使用RDMA快速网络连接. 这是正确的假设吗? 如果使用的是Open MPI，是否需要设置任何特定的标志或命令行说明?
It also states that "all nodes in the pool should have MPI installed (OpenMPI, IntelMPI or any other MPI installer)". From that I understand that as long as you have a pool of A8 / A9 nodes and you are using an image that has any MPI installed, then automatically the nodes will use RDMA fast network connections. Is this a correct assumption? Do we need to set any specific flags or command line instructions if we are using Open MPI?
我要问的原因是，在官方文档(https://docs.microsoft.com/zh-cn/azure/virtual-machines/linux/classic/rdma-cluster)中，仅特别提及了英特尔MPI而不是OpenMPI.
The reason I am asking is that in the official documentation (https://docs.microsoft.com/en-us/azure/virtual-machines/linux/classic/rdma-cluster) only Intel MPI is very specifically mentioned and not OpenMPI.
干杯
卡洛斯
推荐答案
卡洛斯，
不幸的是，该TechNet文章不正确. RDMA对具有infiniband的Azure VM的MPI支持不支持所有MPI堆栈.截至2018年1月23日，当前支持以下功能:
Unfortunately, that TechNet article is incorrect. MPI on RDMA support for Azure VMs with infiniband do not support all MPI stacks. The following are supported, currently, as of 2018-01-23:
- Linux:英特尔MPI
- Windows:MS-MPI
RDMA/Infiniband当前不支持其他MPI堆栈，例如OpenMPI或MPICH.
Other MPI stacks such as OpenMPI or MPICH are currently not supported over RDMA/Infiniband.
干杯
弗雷德
这篇关于在批处理作业中利用带有OpenMPI的RDMA网络的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
