[ewg] OFA EWG Meeting: Monday, Apr, 10, 2017, 09:00 AM US Pacific Time (12pm EST) - Minutes

[Davis, Arlin R]

Jeff, thanks for the reply. Let’s discuss what is driving this before jumping.

David/Rupert,

Can you comment on the need for NFS-rdma; is it a requirement for interop program?
Are there other hard requirements driving the need for NFS-rdma in OFED 4.8?

-arlin

NFS-rdma backports – deferred until OFED 4.8-1 (fall 2017?).

§  IBM would prefer NFS-rdma support in 4.8

§  What is confidence level for NFS-rdma in 4.8-1?

First, it's nice to know that someone wants NFS-rdma for 4.8. When I asked Susan and others at the OFA workshop,  they couldn't name anyone who uses it (except possibly us (NASA) - unfortunately we don't despite my suggestions, although that may soon change).

As far as getting the backports done, I got as far as getting a successful build on both SLES12SP2 and RHEL7.3. Unfortunately, when I tried to run connectathon04 against it, only half the tests passed. After several discussions at OFA workshop with Chuck Lever (upstream NFS-rdma author) and Woody, I decided that the best way to proceed was to start with a base NFS-rdma source earlier than 4.8 (e.g., perhaps 4.5). Then if I get that to build and pass tests, start adding Chuck's patches incrementally if they fix bugs or add important features. Then get code building and tested as patches are added. One big problem with NFS-rdma is that the patches potentially touch all of NFS, and it would not be desirable to backport all of NFS since the distro's have done extensive QA, and NFS is such a critical part of the system.

With that said, I can resume my backport efforts, and see how far I get. Given the amount of work involved, I'm not sure if I can make the May 15 release, but I can try. What do you think?



Thanks,

-jeff


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20170410/aab97297/attachment.html>