[SRP] Feb 1 Meeting notes

Simpson, Cris cris.simpson at intel.com
Fri Feb 1 10:22:24 PST 2002


* From the T10 Reflector (t10 at t10.org), posted by:
* "Simpson, Cris" <cris.simpson at intel.com>
*

Attendees
 
Chuck Gibson    Adaptec
John Carrier	Adaptec (late, silent)
Rob Elliott 	Compaq
Bob Griswold	Crossroads
George Penokie	IBM
Cris Simpson 	Intel
Nathan Obr		Microsoft
Rob Haydt		Microsoft
Ed Gardner		Ophidian
Gerald Mauer    QLogic



Old Business_____________________________________________________

1. Expressing SRP IB ServiceID extensions in IPv6-like :: format
     Response from Marjorie: no interest - DROP - unanimous agreement


2. Review 02-064r2, Ed's SE notification proposal 
     ftp://ftp.t10.org/t10/document.02/02-064r2.pdf
     Ed will clarify where to find sense of bits for Table 6.
     LB resolution will ref 02-064r3.  

3. 02-004 status
     Microsoft expressed concerns about changes to model section.
     WG will review direction at next meeting.     

4. February 8 meeting - conflicts w/T11
     Cancel.

New Business______________________________________________________ 

Discussion of open technical comments

1. HP #1 - Persistent port identifiers
    State that SRP port identifiers have the properties of names (see SAM-2:
persistence, world-wide unique in context of SRP).

>> Add table in Annex A for port name, identifer, etc. SAM mappings.
   Expectation was for persistence already, text will make explicit.   
  

2. HP Page 25 Line 32
    The port identifier fields, at 16B, are too small to carry identifiers
as used by iSCSI. This may prove problematical as we attempt to merge iSCSI
and SRP for use with iWARP.

>>   Statement is true, no change needed. 


3. IBM #50
  (At disconnect) "Discard any subsequent messages received on that RDMA
channel."
   I recommend adding into this list a statement that other SCSI related
   parameters (e.g., mode pages, logs) not be affected by the disconnect. 
   This should avoid the hole the FC has dug for itself in this area.

>>   YES!


4. IBM #57 

  "The following list describes all circumstances where an SRP response
shall not be returned for an SRP request:
    a) If an SRP request conveys a SCSI command and SAM-2 specifies that
STATUS is not returned for that command, then an SRP response shall not be
returned for the SRP request;"

"I am not aware of a SCSI command that specifies that status not be
returned. If there is such a thing then an e.g., would be helpful. If there
is no such thing then this item should be deleted."

>>  The fact that status is not returned is not an attribute of the command,
but of what happened to it during processing. So if Task Mgmt, or error
occured, no status would be returned.  Will reword. (See SAM-2 5.1 for
verbage) 


5. IBM #65 - Removing non-CMDs from flow control
The only SRP request that even needs to have multiple outstanding
requests in [is?] the command. All others should not be streamed but 
should be interlocked and some should be allowed to occur at any 
time. This all needs to be looked at to make sure the
design point is what we really want.

>> Cris: This flow control is really buffer management.  
   Ed suggests changing term from "flow control" to "buffer management".  
   GOP likes "target port buffer management". - (General agreement) 
   George is not concerned about TaskMgmt (essentially the same as CMDs).
   Devolves to: "The initiator may send SRP_I_LOGOUT_REQ notwithstanding the

    value of RequestLimit."  
   Ed, RobE see this as a layering violation.  Possible to disconnect, less
   gracefully, w/o I_LOGOUT. 


6. intel0173 -
Change the value of IB IOControllerProfile:Send Message Depth from Reserved
to be the Maximum Initiator Request Limit.  This allows initiators to
efficiently allocate buffers.
   
   No - too hard to spec this value in a dynamic environment.   

7. intel0174
IB IOControllerProfile:RDMA Read Depth (reserved -> Maximum IOC-issued RDMA
depth). Allows inits to efficiently allocate RDMA resources

   Agreed.


8. intel0175 -
Send Message Size (rsvd -> MAXIMUM INITIATOR TO TARGET IU SIZE) 
    Eliminates need to guess this value

   Since IOC could be used by multiple protocols, needs to say 
   "Shall be at least as big as..."


_______________________________________________________________

Next meeting:

Feb 15 9000-1100 PST

Numbers: 1-888-316-5901,  1-617-801-9781   
Passcode: 2299.1288


--
Cris Simpson                                          503.712.4333         
Intel EPG/ACD Architecture                           Hillsboro, OR   
PGP Fingerprint:0DA0 418E A27B 0B76 5F02  3DD4 0546 6D13 F88A 1E60
*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org




More information about the T10 mailing list