Re: NOP-IN question

On Sep 21, 2007, at 11:40 AM, Ken Craig wrote:

> During my Target development I convinced
> myself that somewhere in the RFC I read
> that only one Target originated NOP IN
> was allowed on the wire at a time per
> Connection.  However I can no longer find
> anything like that in the latest RFC or
> addendums.  I would like to know if I was
> dreaming or can someone point me to a
> place somewhere where the statement is
> made.
>
> Thanks,
> Ken Craig
>
>

Section 8.1 of the iSER draft discusses the need to regulate  
Unexpected messages
sent over an iSER connection, and defines a  
'MaxOutstandingUnexpectedPDUs'
key to regulate this behavior.

An RDMA receiver (iSER or not) must pre-post untagged buffers. If it  
fails to do so
the connection is torn down when the N+1st untagged message shows up.  
How
it does so is left to each ULP. The 'MaxOutstandingUnexpectedPDUs'  
key enables
iSER receivers to safely provision their untagged buffer pool (in  
combination with
the existing logic that tells them how many expected PDUs to allow for).

With plain iSCSI there is a presumption that an excessive number of  
unexpected
messages will *stall* the connection, but not tear it down. Therefore  
there is no
need for any explicit requirements on avoiding this situation.


_______________________________________________
Ips mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ips