|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: iscsi : Review Feedback on Rev 11-94.Santosh, Let me address just this one. Background: Rob Elliott, who is one of this WG contributors as well, is working on a T10 proposal that some of us are collaborating on. The commonly agreed design principle among this team of participants is that it's best if the transport protocols do not individually discuss the SCSI object clearing actions, but rather let SPC-3 do it on a generic "nexus loss" notification event (which is left for each transport to define). In keeping with that goal, iSCSI now groups all SCSI objects into appendix F.2 with the hope that once T10 approves Rob's proposal into SPC-3, the table in this section can be done away with. Now, onto your questions: > Is the "I_T nexus loss" notification being advocated for iscsi targets or > initiators or both ? Both, the text doesn't (intentionally) mention the "role" aspect at all - "iSCSI Layer provides the SCSI layer with the....". >since session > closure occurs upon the ULP's explicit request to close. Hence, the ULP is > aware of the loss of nexus. Not on an Async-solicited single connection session closure. The simplifying design principle I used is to generate a notification whenever the iSCSI session enters the FREE state. I think it's much simpler for iSCSI implementations to always generate a notification upon entering the FREE state. > Also, can we avoid clearing mode pages in case of session re-instatement ? > (case a). In case of ErrorRecoveryLevel=0 initiators, on an iscsi error, > the initiator will log out and re-login, causing a session re-instatement. First off, a "log out and re-login" will not cause a session reinstatement. A connection failure may. Secondly, it appears that you're suggesting a special case for clearing mode pages for a (iSCSI-specific) type of "I_T nexus loss". This is not possible once the SCSI object clearing effects are taken into SPC-3. Having transport-specific "I_T nexus loss" notifications defeats the purpose of having one "I_T nexus loss" event that SCSI ULPs deal with in a consistent way. Thirdly, what an initiator considers as "session reinstatement" may actually be a new session establishment effort on the target if it already timed out the previous session instance - since the network latency in transporting the reinstatement Login is not predictable. IOW, initiators can not make any assumptions about the mode pages in any case. Lastly, the iSCSI session reaches the FREE state on either end on a session reinstatement. Again, it's best to always perform the same actions on reaching FREE as mentioned earlier - no special-casing. Thanks. -- Mallikarjun Mallikarjun Chadalapaka Networked Storage Architecture Network Storage Solutions Organization Hewlett-Packard MS 5668 Roseville CA 95747 cbm@rose.hp.com > 8) Section 4.3.5.1 [TECHNICAL] > ------------------------------ > Is the "I_T nexus loss" notification being advocated for iscsi targets or > initiators or both ? If it is being advocated for the initiator, then, such > notification is not required upon session closure (case b), since session > closure occurs upon the ULP's explicit request to close. Hence, the ULP is > aware of the loss of nexus. > > Also, can we avoid clearing mode pages in case of session re-instatement ? > (case a). In case of ErrorRecoveryLevel=0 initiators, on an iscsi error, > the initiator will log out and re-login, causing a session re-instatement. > > In this case, it is un-necessary to clear the mode page settings, since it > causes extra work in the initiator for no reason. > (need to apply flow control, notify ULP, re-negotiate mode pages, resume I/Os). > In this case, the nexus has not really been lost. It has been re-instated. > Hence, in this case, negotiated mode pages should not be cleared. >
Home Last updated: Thu Apr 11 17:18:21 2002 9610 messages in chronological order |