From: Michael Krause [mailto:krause@...] ... You're assuming an implementation in which RIDs are assigned TO the NIC and disallowing one in which RIDs...
373
Caitlin Bestler
caitikadega
Nov 1, 2001 4:17 pm
... Allowing the NICs full freedom to choose RID values will make load balancing virtual IP addresses across multiple NICs very difficult. It will also make...
374
Renato Recio
recio@...
Nov 1, 2001 4:19 pm
Caitlin, I agree with your comment. I'll make the change. Though... if we were supporting atomics, the data source buffers would also be indeterminate :{). ...
375
Michael Krause
krause@...
Nov 1, 2001 4:33 pm
... More complete response than what I was writing but in agreement with what is stated here. Nothing precludes a NIC from generating its own RIDs - strictly...
376
Hilland, Jeff (Sys Ho...
j2hilland
Nov 1, 2001 4:34 pm
From the protocol standpoint, I believe Ed Gardner said it best so far. If I may rephrase, RID "scope" is relative only to that individual transaction being...
377
Caitlin Bestler
caitikadega
Nov 1, 2001 4:42 pm
I agree that stating nothing works. It probably should be a very explicit nothing however. Even to the point that the holder of a RID/Steering Tag should not...
378
Teisberg, Robert
rteisberg
Nov 1, 2001 4:59 pm
I agree. The appropriate "protocol"; for communicating information about RID/Steering Tag scope involves RFPs and programmers guides and nothing more. ... ...
379
Michael Krause
krause@...
Nov 1, 2001 4:59 pm
... The implementation guide is going to be treated by many of us a de facto standard for product development. I don't have a problem with the protocol spec...
380
Stephen Bailey
steph@...
Nov 1, 2001 8:53 pm
... I'd prefer to phrase this differently. The wire protocol specification represents a highly packed representation of normative choices, selected upon the...
381
Kanevsky, Arkady
arkady@...
Nov 1, 2001 11:46 pm
I have one concern. I am looking into "one time use memory windows" or more generally ability for the Memory Window user to specify to the provider of the...
382
Caitlin Bestler
caitikadega
Nov 2, 2001 12:20 am
The fact that a RID/STag is one time use only is covered by the statement that a RID/STag *shall not* be used in any many that is not authorized by the ...
383
Bernard Aboba
aboba@...
Nov 2, 2001 1:27 am
Starting next week, we will be having weekly "Case for RDMA" document meetings on Wednesdays, from 12 noon to 1 PM PST. Here is the conference call info: Date:...
384
Madhusudhan Talluri
madhusudhan.talluri@...
Nov 2, 2001 1:39 am
... In the case I am using a TCP LLP connection. Can the ULP send traffic using non-iWARP data operations, like doing direct socket writes or is a TCP...
385
Caitlin Bestler
caitikadega
Nov 2, 2001 1:46 pm
... iWARP mode and non-iWARP modes are mutually exclusive. In previous posts concerning connection establishment I covered some of the reasons. The most...
386
Edward A. Gardner
edwardallang...
Nov 2, 2001 2:05 pm
... What do you have in mind when you say "one time use"? Anything that would imply iWARP protocol support? The following example comes to mind, is this at...
387
Culley, Paul
prculley
Nov 2, 2001 2:39 pm
Regarding the mixing of iWarp segments and stream data: While you are technically correct that stream data -could- look like an iWarp segment, I regard this as...
388
Culley, Paul
prculley
Nov 2, 2001 2:40 pm
As noted earlier, enabling the "one time use" DOES require a bit of support from the wire protocol; specifically a way for the target to know when the data...
389
Caitlin Bestler
caitikadega
Nov 2, 2001 2:49 pm
A "last" bit could make things easier. However it is not necessary. The automatic de-registration could take place on the next receive over the iWARP channel. ...
390
Michael Krause
krause@...
Nov 2, 2001 3:09 pm
... Agreed Mike...
391
Michael Krause
krause@...
Nov 2, 2001 3:09 pm
... Relying on a wire event to occur to signal a previous event creates a window with a set of undesirable implications. As you note, it isn't hard, we have...
392
Jim Pinkerton
jpink@...
Nov 2, 2001 3:25 pm
Comments below... ... a ... submit ... piece ... operation ... question ... due ... SHALL ... or ... for ... When we came up with the initial proposal for...
393
Jim Pinkerton
jpink@...
Nov 2, 2001 3:25 pm
On the "anonymous buffer" issue, I think the key concern I had when raising the issue at the face-to-face was that some people view anonymous buffers as being...
394
Jim Pinkerton
jpink@...
Nov 2, 2001 3:25 pm
As designated maintainer of the definitions section of the wire spec, I've been asked to update the terms to our current state of thought, and direct review of...
395
Caitlin Bestler
caitikadega
Nov 2, 2001 3:36 pm
Some ULPs, including DAFS, will require the ability to negotiate the size of the SEND/RECV buffers themselves. Specifically they may have protocol state...
396
Edward A. Gardner
edwardallang...
Nov 2, 2001 4:23 pm
... Along the lines of earlier discussion on RID scope, the buffer queuing model is outside the scope of the wire protocol and therefore outside the scope of ...
397
Edward A. Gardner
edwardallang...
Nov 2, 2001 4:23 pm
... Having already been bitten by "segment" confusion, I'm particularly sensitive to it. With that said, I can live with "iWARP segment", so long as we are...
398
Williams, Jim
jimwgiganet
Nov 2, 2001 4:42 pm
... If anonymous buffers are queued on a per NIC basis, that makes it impossible in general for a ULP to make sure that messages are not dropped due to no...
399
Jim Pinkerton
jpink@...
Nov 2, 2001 10:25 pm
I'm just trying to figure out where consensus lies, and really don't feel terribly strongly one way or the other. That said, comments below... Jim ... design, ...
400
kahmin.teh@...
Nov 3, 2001 1:57 am
unsubscribe rdma@yahoogroups.com...
401
Michael Krause
krause@...
Nov 3, 2001 2:51 am
... I had really hoped that the f2f could run concurrent, i.e. the day of, before or after the workgroup meets so that those of who travel only have to be...
