Dear Colleagues: I've recently submitted a draft including a different "No-Data-Touch"(NDT) framer. Unlike the current proposed NDT framer that has a "mistake"...
825
Stephen Bailey
stephen_w_ba...
Feb 2, 2002 8:37 pm
I added the `notifying write' bit, which Caitlin so astutely pointed out was missing from my RDMA thingy. It's in the same place: ...
826
Julian Satran
julian_satran@...
Feb 7, 2002 2:10 am
Martin, I am glad you like it. The salt is a random number (a RN generator generates a new one at every PDU). That is why it is hard to guess. PDU headers are...
827
Culley, Paul
prculley
Feb 7, 2002 2:09 pm
I am having a hard time understanding the advantage of the "salt", as opposed to a longer framing key. Both would seem to have the same probability of false...
828
Julian Satran
julian_satran@...
Feb 7, 2002 8:28 pm
And in addition even if the frame key is random it does not change for the life of the stream - and that makes it impractical. Julo "Culley, Paul" ...
829
Julian Satran
julian_satran@...
Feb 7, 2002 8:28 pm
Paul, The salt is a random number that is changing on every PDU. That changes radically the false detection probability in face on an infinite stream form...
830
Stephen Bailey
stephen_w_ba...
Feb 8, 2002 3:17 am
Julian, I was also initially excited by your proposal, but I've come to a similar conclusion to Paul. The key question is what property is probability in the...
831
Julian Satran
julian_satran@...
Feb 8, 2002 4:48 pm
yes - except that unlike a dictionary builder I do not attempt to guarantee that it is an "only once" word :-) julo Lloyd Wood <l.wood@... To:...
832
Julian Satran
julian_satran@...
Feb 9, 2002 4:54 am
Steph, The mathematics behind the proposal are simple. You have two random numbers "against" a finite-length PDU. In you original proposal you have single...
833
Stephen Bailey
stephen_w_ba...
Feb 9, 2002 4:35 pm
For a proper safety analysis, you can not assume that the the contents of the the stream (or the PDU) are randomly distributed. You have to assume the stream...
834
Julian Satran
julian_satran@...
Feb 10, 2002 7:21 pm
Leonid, The probability is not 0 - that is true. But for the form used in the ULP framing draft it is in theory 1 and that is unacceptable. The new proposal is...
835
John Hufferd
hufferd@...
Feb 11, 2002 1:27 am
Julian, I do not think any of that matters, since you are not trying to determine the probability of a False Negative. Only the probability of a False ...
836
Culley, Paul
prculley
Feb 11, 2002 2:19 pm
While I agree that the probability approaches 1 with an infinite stream, I believe that this is true for both your approach and the TUF approach, it makes no...
837
Julian Satran
julian_satran@...
Feb 11, 2002 5:39 pm
John - it is all about false positive. Julo John Hufferd@IBMUS 11-02-02 03:26 To: Julian Satran/Haifa/IBM@IBMIL cc: "Leonid Grossman"...
838
Julian Satran
julian_satran@...
Feb 11, 2002 7:07 pm
Paul, I strongly disagree. The probability with my proposal is not dependent on the stream length only on the PDU length. You don't have to test for infinite...
839
Stephen Bailey
stephen_w_ba...
Feb 11, 2002 10:41 pm
John, Thanks for the exhausting analysis. As Paul points out, you do not want to scan the received segment to recover synchronization with either of these...
840
Douglas Otis
doug_otis
Feb 12, 2002 12:13 am
Steph, Schemes exclusively supporting layers prior to TCP, used in conjunction with application designators for placement, are advocating a TCP/application ...
841
John Hufferd
hufferd@...
Feb 12, 2002 1:04 am
Steph, How small do you think the segmentation will be, and will they generally be on a word boundary? The value I used for 361 or 362 (lets call it the ...
842
Julian Satran
julian_satran@...
Feb 12, 2002 7:00 am
Steph, You fail to see that a good synchronization scheme should work with or without a TCP alignment scheme. If the instream alignment can be infered by the...
843
Douglas Otis
doug_otis
Feb 12, 2002 12:46 pm
Julian, An application using DDP must continuously communicate with a steering layer added to the transport. Placement is an intimate aspect of transport ...
844
joea.bailey@...
Feb 12, 2002 5:30 pm
Is there a web site or contact?...
845
Stephen Bailey
stephen_w_ba...
Feb 12, 2002 8:19 pm
... I see it clearly. However, this seems to be a case where we can't have everything. If we want to touch the data we can have a synchronization technique...
846
WENDT,JIM (HP-Rosevil...
jwendt999
Feb 13, 2002 7:16 pm
... Not to muddy the waters, but isn't an "aligning FIM" possible (not that anyone would want to do it), where ULP PDUs (possibly chunked) are aligned into TCP...
847
WENDT,JIM (HP-Rosevil...
jwendt999
Feb 13, 2002 7:17 pm
The flexibility to synchronize independent of transport segment alignment would seem to come at the costs of a significant increase in receiver complexity in ...
848
Julian Satran
julian_satran@...
Feb 13, 2002 8:39 pm
For non-generic hardware DDP and alignment are not synomins. Neither are they necessarily for generic hardware. Julo "WENDT,JIM (HP-Roseville,ex1)"...
849
John Hufferd
hufferd@...
Feb 13, 2002 11:23 pm
Paul, Sorry I did not reply sooner, I just focused on your note, and found that you were suggesting the same thing I was asking myself. That is, since the...
850
THALER,PAT (A-Rosevil...
pat_thaler@...
Feb 15, 2002 12:13 am
Jim, I don't understand what you are suggesting. It seems like an "aligning FIM" approach could only work for a ULP that had fixed length PDUs - which would ...
851
ACE SEO
sales@...
Feb 16, 2002 11:31 pm
ACE Search Engine Optimization ACE Search Engine Optimization GRAND LAUNCH of ACE Search Engine Optimization Go to our site click here! ACE gets you to the TOP...
852
allyn
allyn@...
Feb 21, 2002 10:38 pm
There is a new problem statement for RDMA over IP. Until it is available in the internet drafts directory, you can get it at ...
853
Douglas Otis
doug_otis
Feb 21, 2002 11:43 pm
Allyn, I would not characterize RDMA as being synonymous with Direct Data Placement. Direct Data Placement would be a portion of the function performed by...
