The Fujitsu Channel Bonding solution has been submitted to the community tree under patch:
http://review.whamcloud.com/#/c/14625/. This solution enables the bonding of multiple
OFED supported interfaces in active-active mode. So, you could have 2 or more IB
interfaces over which LNet messages are sent round robin. If a connection goes down, a
background process will periodically try to re-establish it while messages continue to be
sent over remaining interfaces in the bond.
Many community members are showing interest in seeing this patch land for the 2.8 release
(feature freeze is June 30th) as it brings performance and additional high availability
features to Lustre. This feature is not listed on the 2.8 release page
(
http://wiki.lustre.org/Release_2.8.0), but the desire to land it is high.
To help meet the June 30th deadline, we need more community participation in reviewing and
testing this patch. Without that, I fear this feature may not make the 2.8 release.
An additional patch which goes with it is:
http://review.whamcloud.com/#/c/15170/. This
adds support to the Dynamic LNet Config feature (landed in 2.7) to configure the bond.
This is important to having one universal way to configure channel bonding right from the
beginning.
It would be great if we can get reviewers to look at these two patches as soon as
possible.
Doug