[ofiwg] DS/DA - kernel maintainer slide deck

Bernard Metzler BMT at zurich.ibm.com
Mon Feb 1 04:31:16 PST 2016


Slide 12 - this is a nice figure.

I suggest to split the arrow between
[SRP, iSer, NFSoRDMA...] and [kverbs] if it shall touch
- and not bypass - kfabric. Of course, as of today,
there is a direct link between these two boxes and no
need to talk to kfabric to make iSER happen. So we may
end up with a slightly shorter [kfabric] box to give
to its left way for a direct link between [iSER] and
[kverbs] as well.

Regarding your question for me - I would hope to see a
common interface to byte-access storage, irrespective if
remote or local. For that, I would expect [kfabric] to
appear also on the left side, at the same level as on
the right. It then remains up to the use case if the
application would like to be aware of data locality,
but that discussion is imho orthogonal to storage access
provision as shown in the figure.
All this of course shall not prevent other path's to NVM
as currently depicted on the left side of slide 12 - but
they do not relate to kfabrics or kverbs.

Thanks & best regards,

> From: Paul Grun <grun at cray.com>
> To: "'ofiwg at lists.openfabrics.org'" <ofiwg at lists.openfabrics.org>
> Date: 01/30/2016 02:40 AM
> Subject: [ofiwg] DS/DA - kernel maintainer slide deck
> Sent by: ofiwg-bounces at lists.openfabrics.org
> I updated the slides based on Tuesday’s discussion.  I also began
> thinning out the selection of backup and bone pile slides.  I think
> we are close.
> Two remaining open questions:
> Bernard – referring to slide 12, do we need to say something here
> about the desirability of a single method for accessing NVM
> regardless of whether it is local or remote?
> Frank – Do we want to add back a slide illustrating the HA use case
> you mentioned early in the discussion?
> -Paul
> Cray Inc.
> Office:    (503) 620-8757
> Mobile:  (503) 703-5382
>  [attachment "kfabric-maintainer discussion__2016_0129.pptx" deleted
> by Bernard Metzler/Zurich/IBM]
> ofiwg mailing list
> ofiwg at lists.openfabrics.org
> http://lists.openfabrics.org/mailman/listinfo/ofiwg

More information about the ofiwg mailing list