SORT BY:

LIST ORDER
THREAD
AUTHOR
SUBJECT


SEARCH

IPS HOME


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

    Re: aborting an out of sequence cmdSN



    Julian,
    
    I understand that nothing in the current draft prevents usage of
    connection allegiance for an abort task to flush the stale PDUs of a
    command. 
    
    However, as a target implementor, if such a guaranteed mechanism is not
    mandated to flush stale PDUs of a command, targets will need to build
    safeguards against possible stale PDUs from a command. Any such scheme
    would cause targets to retire resources until a safe period (based on a
    best guess !) has elapsed. The targets CANNOT depend on initiators
    deploying this scheme in the absence of the spec mandating this and do
    not have a guarantee that stale PDUs will not arrive after the Abort
    Task has cleaned up the command at the target end.
    
    Keeping in mind the above, I am asking that iSCSI mandate that the Abort
    Task be sent on the same connection as the SCSI command being aborted.
    In addition, [and this has been asked earlier], the iSCSI draft lacks a
    section on I/O timeout handling under the error recovery section. Some
    description must be provided on the actions an initiator should take on
    an I/O timeout. (or timeout of other iSCSI PDUs).
    
    For instance, should non-SCSI PDUs be timed ? (login, text, etc). If so,
    what is their timeout values. What actions should an initiator take if
    they time out. How is the task tag for that non-SCSI PDU cleaned up,
    ensuring no stale PDUs are going to arrive for that task tag.
    
    iSCSI Error Recovery descriptions MUST address task timeout handling. 
    
    - Santosh
    
    
    julian_satran@il.ibm.com wrote:
    > 
    > First the current draft does not stop you from doing what you want - it
    > just does not mandate it.
    > Second with CmdSN now you have a better chance - before handing the command
    > over to SCSI task management
    > mark all the relevant commands in the iSCSI queue as non-deliverable to a
    > LUN or all LUNs or keep a stack of barriers for active aborts (I assume
    > that this is a reasonably low number) to the same effect.
    > 
    > Julo
    > 
    > Santosh Rao <santoshr@cup.hp.com> on 18/04/2001 01:10:17
    > 
    > Please respond to Santosh Rao <santoshr@cup.hp.com>
    > 
    > To:   Sandeep Joshi <sandeepj@research.bell-labs.com>
    > cc:   ips@ece.cmu.edu
    > Subject:  Re: aborting an out of sequence cmdSN
    > 
    > I'd think option 1 is the simplest (with the caveat that the task mgmt
    > PDU referred to is the Abort Task.) and only impacts the affected
    > command/task.
    > 
    > Pierre Labat and I have asked for this 4 months ago. (See :
    > http://ips.pdl.cs.cmu.edu/mail/msg02958.html). The concept of connection
    > allegiance should be extended to include the Abort Task. Also,
    > connection allegiance should apply to the task (which spans multiple
    > commands in the case of linked commands.), allowing for a deterministic
    > clean up of stale PDUs of the task through the use of Abort Task.
    > 
    > - Santosh
    > 
    > Sandeep Joshi wrote:
    > 
    > >
    > > So our options for abort_task boil down to..
    > > (1) use connection allegiance for TASK MGMT PDU.
    > > (2) reject all commands prior to cmdSN of TASK MGMT PDU.
    > > (3) cmdSN of original task is sent with TASK MGMT PDU and
    > >     target at the iSCSI layer keeps state.
    > > (4) iSCSI initiator retains state for deleted tasks to ensure
    > >     that R2T/Scsi Responses are appropriately handled.
    >  - santoshr.vcf
    begin:vcard 
    n:Rao;Santosh 
    tel;work:408-447-3751
    x-mozilla-html:FALSE
    org:Hewlett Packard, Cupertino.;SISL
    adr:;;19420, Homestead Road, M\S 43LN,	;Cupertino.;CA.;95014.;USA.
    version:2.1
    email;internet:santoshr@cup.hp.com
    title:Software Design Engineer
    x-mozilla-cpt:;21088
    fn:Santosh Rao
    end:vcard
    


Home

Last updated: Tue Sep 04 01:05:00 2001
6315 messages in chronological order