|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] RE: SCSI Format/Copy CDB supportThat would be a mistake. It may affect commands intended to be dropped. If the window has moved past CmdSN you DO NOT reissue the command (that is clear in the text). Julo
Julo, We only need to ADD a single line to section 2.2.2.1 as follows: "CmdSN must change if the ExpCmdSN/MaxCmdSN window has moved." Section 2.2.2.1 is saying: Amir From David Black: The CmdSN window should take care of the "long time" issue - once the Window (ExpCmdSN/MaxCmdSN) has moved past a particular CmdSN, the initiator knows that the target successfully received the command, and reissuing the command with the same CmdSN would be an error. Julian - please add a suitable clarification to the next version of the spec. Thanks, --David -----Original Message-----
David, I agree. Lets clarify it in the spec. We also must make sure that iSCSI doesn't reissue a command due to connection drop that happened a "long time" after the command was originally issued. Amir -----Original Message----- From: Black_David@emc.com [mailto:Black_David@emc.com] Sent: Saturday, January 26, 2002 12:32 AM To: amir@astutenetworks.com Cc: ips@ece.cmu.edu Subject: RE: SCSI Format/Copy CDB support Ah, I see the confusion. In that section, "reissued" refers to reissuing a command due to holes in the command sequence (e.g., digest failure, connection drop). In the scenario of interest, the command is not being reissued by iSCSI - SCSI is aborting the old command based on the task tag and issuing a new command which will get a new CmdSN (iSCSI will probably have long since forgotten the CmdSN assigned to the original command). A sentence or two to clarify this wouldn't hurt. --David > -----Original Message----- > From: Amir Shalit [mailto:amir@astutenetworks.com] > Sent: Friday, January 25, 2002 7:03 PM > To: Julian Satran > Cc: ips@ece. cmu. edu (E-mail); owner-ips@ece.cmu.edu > Subject: RE: SCSI Format/Copy CDB support > > > Julo, > > I was talking about initiator aborting the command > ,say two hours after it was sent, and than reissue > the same command with the old CmdSN (according to > the spec). > > I agree with you that the spec covers all other cases > since once a command is delivered, CmdSN doesn't > matter much and ITT or TTT are used to match the > task. > > Amir > > -----Original Message----- > From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of > Julian Satran > Sent: Friday, January 25, 2002 2:14 PM > To: Amir Shalit > Cc: ips@ece. cmu. edu (E-mail); Julian Satran; owner-ips@ece.cmu.edu > Subject: RE: SCSI Format/Copy CDB support > > > > Amir, > CmdSN is completely secondary in abort task. > The main element is the Initiator Task Tag. > If ITT is not found at abort then the target, under some > cirscumstances, > will look for a hole at CmdSN to fill it or will answer > negatively the Task > Abort. > > Julo > > > > "Amir Shalit" > <amir@astutenet To: Julian > Satran/Haifa/IBM@IBMIL > works.com> cc: "ips@ece. cmu. edu > \(E-mail\)" <ips@ece.cmu.edu>, > <owner-ips@ece.cmu.edu> > 25-01-02 18:49 Subject: RE: > SCSI Format/Copy > CDB support > > > > > > > Julo, > > I agree with you on (2). > > Section 2.2.2.1 is saying: > "A numbered iSCSI request will not change its allocated CmdSN > regardless of the number of times and circumstances in which it is > reissued." > > A long duration SCSI command may be aborted by a task management > command and than reissued by iSCSI with a stale CmdSN. > > Amir > -----Original Message----- > From: owner-ips@ece.cmu.edu > [mailto:owner-ips@ece.cmu.edu]On Behalf Of > Julian Satran > Sent: Friday, January 25, 2002 1:21 AM > To: Amir Shalit > Cc: ips@ece. cmu. edu (E-mail); owner-ips@ece.cmu.edu > Subject: Re: SCSI Format/Copy CDB support > > > Amir, > > The draft says in several places that CmdSN is not > important after the > task gets to SCSI. > On 2 yes you might be right but that is purely an implementation > issue. Format commands do not time-out > at SCSI level and iSCSI has no cause to e "nervous" - nothing is > expected. > > > Julo > > > > > "Amir Shalit" > <amir@astutenetworks.com> To: > "ips@ece. cmu. edu > Sent by: owner-ips@ece.cmu.edu \(E-mail\)" <ips@ece.cmu.edu> > cc: > Subject: > SCSI Format/Copy > 25-01-02 02:10 CDB support > > > > > > > > I see a few problems in implementing the format/copy > commands over > iSCSI > > 1) CmdSN may wrap between command and response (assuming multiple > LUNs) > > 2) iSCSI keepalive NOPs may be required to keep the > connection open > during format > > > > > > >
Home Last updated: Mon Jan 28 15:18:04 2002 8521 messages in chronological order |