Re: [Ips] I-D ACTION:draft-ietf-ips-iser-00.txt

Caitlin,

Section 7.3.6 clause 1 describes the behavior of the iSER layer at the 
target as it invokes the RDMAP layer to send the RDMA Read Request to the 
initiator.  Your comments, however, seem to be directed to the RDMA Read 
Reply as handled by the RDMAP layer at the initiator.  If that is the 
case, then I agree with you that for the iSER layer at the initiator, "it 
is also not appropriate for iSER to presume which choice has been made" as 
far as the Data Sink STag in the RDMA Read Request is concerned since RDMA 
Read Replies are handled by the RDMAP layer.

Mike
To:     Mike Ko <[email protected]>
cc:     [email protected], [email protected] 
Subject:        Re: [Ips] I-D ACTION:draft-ietf-ips-iser-00.txt




On Sep 15, 2004, at 8:00 PM, Mike Ko wrote:

> Caitlin,
>
> Section 7.3.6 clause 1 states that:
>
> 1.  It MUST ensure a valid local STag for the I/O Buffer and a valid
>     local mapping that associates the Initiator Task Tag (ITT) to
>     the local STag.  This may involve allocating a valid local STag
>     and establishing a local mapping.
>
>
To clarify further. There is an inherent difference in the wire protocol
between STag used for RDMA Writes and STags that are used for
RDMA Read Replies.

STags used by RDMA Writes are explicitly distributed by the ULP
and their use is selected by the ULP.  Therefore *any* local
interface MUST provide the ability to the consumer to create
an STag matching a specific buffer (with a subset of the authority
the consumer already had to access that buffer).

The STag used in an RDMA Read Reply MUST be a simple
echo of the value supplied in the request. It is NOT exposed
to the ULP. Therefore an RDMAP implementation MAY use
a different algorithm for allocating and tracking STags for
RDMA Reads than it uses for RDMA Writes.  Further, there
are certain security, feature and efficiency benefits for
taking this approach.  The tradeoffs between the "just
like tagged" and "just like untagged" is complex, and not
really appropriate for either the IPS or RDDP -- but it is
also not appropriate for iSER to presume which choice
has been made.