At the last DAT call the issue of the scope of FMR LMR came up. Ne decision was reach and the issue is no pushed to reflector for more feedback. Here is the...
AT the alst meeting we decided to add one more argument for LMR_bind to specify a physical page size. But we do not have the page size for LMR_create. Should...
I got an AI from alst DAT call to propose LMR_state. Here is the first draft: struct dat_lmr_state { DAT_BOOLEAN fmr_support; DAT_BOOLEAN bound; }; ...
At the last DAT call there was a discussion on the LMR_Completion event format. There are 2 interrelated questions. 1. Do Consumer want to get LMR_handle of...
At the last DAT call there was a discussion on the fencing behavior of LMR_bind (FMR). IBTA 1.2 and iWARP do not seem to define a behavior for the FMR ...
Hello, This email message is a notification to let you know that a file has been uploaded to the Files area of the dat-discussions group. File :...
dat-discussions@yahoo...
Sep 2, 2005 6:44 pm
3559
A few thoughts on the issue of FMR completions, and on whether local fencing needs to be exposed. These are inter-related because the critical question is...
Hello, This email message is a notification to let you know that a file has been uploaded to the Files area of the dat-discussions group. File :...
dat-discussions@yahoo...
Sep 2, 2005 7:26 pm
3561
EP_handle is always returned in completion event. The question is should we also return LMR_handle (analogous to RMR_bind RMR_handle). If you recall the...
I think at the end we will merge both completion formats into one. For remite invalidate recv completion will have to return which RMR_context have been...
We will have a DAT Collaborative conference call this Wednesday Sept. 7, 12:00-1:00pm EDT (9:00am-10:00am PDT). Moderator: Arkady Kanevsky Phone: 888-827-8686 ...
1. Do we have any need to support combined operation for Local Invalidate with actual Send? Or an lmr_bind with length 0 is sufficient? Trying to see if we...
I had tentatively added to the DTO_Completion 2 new arguments: DAT_BOOLEAN invalidate_flag; DAT_RMR_CONTEXT rmr_context; These are needed to handle remote...
Text tentatively added to dat_post_recv: The asynchronous successful completion of the posted Receive also indicate whether remote side invalidated an...
Hello, This email message is a notification to let you know that a file has been uploaded to the Files area of the dat-discussions group. File :...
dat-discussions@yahoo...
Sep 6, 2005 7:22 pm
3569
Updated kDAPL spec. The changes: 1. FMR requirements clarification for different page size support for different FMRs 2. added send_with_invalidate 3....
This is more aggressive that I was anticipating - ref the registration caching and the meeting tomorrow. Using FMRs from user mode seems expensive. Caching...
On the 07 sept dat call, there was a discussion on adding a new policy for the evd upcall - here are the outlines of the problem and the suggested change: ... ...
One quick question/thought: With the "ONE SHOT" policy, which rules applies: the DAT Provider SHOULD NOT provide an event, but MAY do so when the event has...
There is also a question on what should happen when "DAT_UPCALL_ONE_SHOT" policy is set by dat_evd_modify_upcall and there is an event on the queue. I think we...
... I think the _second_ course is the preferred one. As you suggested in the call - there can be a bogus event delivered to the consumer (to keep the current...
... I think that when a consumer requests for notification he should get notified for existing events too. If this is not the default way, there should indeed...
Guy, How can the number of events on the queue be provided without dequeueing them? I recall that the reason we only have a binary indicator for more_events in...
Here is a proposal to handle registration of the physical pages of different sizes. As agreed we will only support one physical page size per registration. But...
... in the openib verbs, for instance, you use ib_peek_cq for the dto completions, and the connection events are in the pending list - you can count them. ... ...