Greetings - There was a lively discussion at the last meeting built around options for adapting LNET to kfabric. In the course of the discussion, several interesting points were raised:
* Areas for augmenting the existing kfabric proposal:
* CPU/NUMA affinity
* Improved/simplified memory management
* Automatic connection establishment
* Whether kfabric, as currently proposed, is at the right level of
abstraction for Lustre.
* Current kfabric proposal closely mimics libfabric, which has complex
requirements given its intended usage model.
* Should the existing proposal be augmented with vastly simplified
get/put semantics to more closely mimic the needs of I/O?
The minutes from the last meeting are attached.
Agenda for tomorrow is to discuss these ideas. Do we need to put kfabric
through a new round of requirements gathering?
Regards,
-Paul
Advanced Technology Group
Cray, Inc.
Office - (503) 620 - 8757
Mobile - (503) 703 - 5382
OFWG_DSDA_minutes_2017-01-24.docx
Description: OFWG_DSDA_minutes_2017-01-24.docx
_______________________________________________ ofiwg mailing list [email protected] http://lists.openfabrics.org/mailman/listinfo/ofiwg
