Question about Virtual RDMA support in containers

Hi folks,

I’m trying to understand if the virtual RDMA can be supported in containers which share the single HCA; for that design, I assume in each container it cans till take the benefits bring by RDMA kernel/CPU-bybass feature. I did see SRIOV solution but the artical can be found only gives rough intro without details.

The goal is actually to let containers running in their own vlan can talk to each other (could be located in different physical nodes) with RDMA, so, it can bring flexible deployment (containers) and with better network performance (RDMA) .

Really appreciate for any suggestions.

Huide

Possibly this question need to be post in Mellanox OFED sub-category also? please suggest.

Hello Huide,

Welcome, and thank you for posting your inquiry to the NVIDIA developer forums!

Many end-to-end deployment examples can be found on our Solutions landing page:
https://docs.nvidia.com/networking/display/public/SOL/Solutions+Home

Relevant to your use case, we have 2 examples documented there that walk through configuration of both Docker and LXD containers with RDMA over Infiniband fabrics:

https://docs.nvidia.com/networking/pages/releaseview.action?pageId=15049785

https://docs.nvidia.com/networking/pages/releaseview.action?pageId=15049745

Thanks,
NVIDIA Enterprise Support

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.