SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


    [Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

    iSCSI: Ordered delivery capability notification]


    • To: ips@ece.cmu.edu
    • Subject: iSCSI: Ordered delivery capability notification]
    • From: Pierre Labat <pierre_labat@hp.com>
    • Date: Wed, 10 Jan 2001 17:11:11 -0800
    • Content-Type: multipart/mixed;boundary="------------BA2F059E6BBC901F2631925D"
    • Organization: Hewlett Packard ATM-SISL
    • Sender: owner-ips@ece.cmu.edu

    Hoops! I forgot to add the "iSCSI"  prefix
    


    • To: ips@ece.cmu.edu
    • Subject: Ordered delivery capability notification
    • From: Pierre Labat <pierre_labat@hp.com>
    • Date: Wed, 10 Jan 2001 19:45:14 -0800
    • Content-Transfer-Encoding: 7bit
    • Content-Type: text/plain; charset=us-ascii
    • Organization: Hewlett Packard ATM-SISL
    • Sender: plabat@cup.hp.com
    Julian,
    
    
    Where:
    ======
    
    1.2.2.1 Command numbering
    
    "iSCSI supports ordered command delivery within a session"
    
    "The target iSCSI layer SHOULD deliver the commands to the target
    SCSI layer in the order specified by CmdRN".
    
    
    Problem:
    =======
    An initiator is unable to know if the iSCSI target layer support ordered
    
     command delivery. Hence it cannot trust iSCSI to have ordered delivery.
    
    Hence all the iSCSI ordering mechanism becomes useless because
    it can't be trusted.
    The problem is not that iSCSI target layers can't support ordered
    delivery,
    the problem is that the initiator/application doesn't know anything
    about it.
    
    
    Solution:
    ======
    
    a) Add a Login/Text key
    Ordered:<yes|no>
    
    b) Put the explanation in for the key:
    "If the target answered "yes" and is unable to maintain the order for
      whatever reason, it must notify the initiator and drop the session"
    
    If the target answered "yes", when it receives a  "retried" command
    (retry bit),
    it must be able to retry the command in such a way that the effect
    on the initiator and target device when the command completes
    is the same as if the command would have not needed to be retried.
    This doesn't mean always restart from where left, a READ
    can have all the data re-transmitted because it is transparent
    a completion time.
    
    
    
    




Home

Last updated: Tue Sep 04 01:05:54 2001
6315 messages in chronological order