[ofiwg] New Intel slides from today’s OFI WG discussion

Paul Grun grun at cray.com
Tue Dec 2 14:58:40 PST 2014


I have also moved Bernard Metzler’s presentation from last week’s meeting, entitled byte_io_reqs.pdf, to the same DSDA Presentations<https://www.openfabrics.org/downloads/OFIWG/DSDA%20Presentations/> area.

From: ofiwg-bounces at lists.openfabrics.org [mailto:ofiwg-bounces at lists.openfabrics.org] On Behalf Of Paul Grun
Sent: Tuesday, December 02, 2014 2:46 PM
To: Douglas, Chet R; Reese Faucette (rfaucett); Voigt, Doug; ofiwg at lists.openfabrics.org; Walt Hubis
Subject: Re: [ofiwg] New Intel slides from today’s OFI WG discussion

Thanks Chet.  I have uploaded these to the OFA / OFI WG download site: https://www.openfabrics.org/downloads/OFIWG/
Look under ‘DSDA Presentations<https://www.openfabrics.org/downloads/OFIWG/DSDA%20Presentations/>’.
-Paul

From: Douglas, Chet R [mailto:chet.r.douglas at intel.com]
Sent: Tuesday, December 02, 2014 2:28 PM
To: Reese Faucette (rfaucett); Voigt, Doug; ofiwg at lists.openfabrics.org<mailto:ofiwg at lists.openfabrics.org>; Walt Hubis; Paul Grun
Subject: New Intel slides from today’s OFI WG discussion
Importance: High

Hello OFI WG Members,
I have enclosed a new copy of the slide deck that I shared this morning in the OFI WG.  Please disregard the previous deck and post this to the OFI WG.

Here is what I clarified in the latest slides.  I will bring these changes to everyone’s attention in a future meeting:
-Made it clear that the IIO internal buffers are usually backed by LLC core cache when “allocating-write” flows are utilized
-Removed Send/Receive as a mechanism for clearing out the internal IIO buffers.  RDMA Read or a PCI Read of some sort is required.
-Fixed up the Non-ADR, DDIO Enabled slide to make it clear that standard “allocating write” transactions can be utilized since the LLC backed memory will be flushed by the CLFLUSHOPT instruction.   For the same reason, the RDMA Read is not needed in this case.  The Send message must contain a list of all cache lines that were written by the stream of RDMA Write(s) so that the CLFLUSHOPT can be utilized to flush each cache line.

I will be presenting these slides to the SNIA TWG at a near future date for those that are also involved with that group.  I will also add sequence diagrams to help explain each of these flows.
Thanks,
Chet





___________________
Chet Douglas
Principal Engineer
DCG - Crystal Ridge SW Architecture
Intel Corporation
Office:  (480) 554 2669
Mobile: (520) 241 8038






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofiwg/attachments/20141202/c4fa7870/attachment.html>


More information about the ofiwg mailing list