|[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]|
Mallikarjun & All,As I pointed out in a private discussion with you I never considered in dept the response-fence (beyond what is needed for TM). It is a serious issue for any "transaction type" operations to storage - but the last interface that had all the pieces correctly aligned was the 360 channel. SCSI had taken (traditionally) the positions that sequencing and barriers are rarely needed and when needed applications should provide them. And that is exactly what all of the relevant applications do (all transaction monitors, databases etc.).
The way all the application handle cases when the start of some SCSI command has to follow the end of a previous set of operations is to just wait for the end of the operations before issuing the "dependent commands".
Is this solution optimal? As you probably observed already this solution implies that all the queues in the "stack" (at initiator, transport and target) get empty before the dependent commands can be issued and that is bad for performance.
Can fencing (bad name BTW because it is in widespread use for defect isolation) improve this? Yes (by keeping the pipes full!) provided it is associated with rigorous exception handling and the later is a SCSI issue (exceptions will require emptying the queues). With most applications being distributed I doubt also that fencing for a single I_T nexus is interesting and coordinating fencing and exception handling over several I_Ts is a complex exercise.
And as transport alone brings no value to the whole issue - why should we go through the exercise of defining a solution before SCSI has defined one and we are confident that it is good enough and has value? Again I agree that we a mechanism for TM but a good enough mechanism for TM is already in and it gets now even better even without resorting to a generalized fencing/barrier support.
Regards, Julo Mallikarjun C. wrote:
Trying again after bouncing (>50KB), tail is now clipped.... ----- Forwarded Message ---- From: Mallikarjun C. <cb_mallikarjun@xxxxxxxxx> To: ips@xxxxxxxx Sent: Friday, January 5, 2007 2:40:56 PM Subject: Re: Response Fence Flag Hi John,As you correctly point out, there are two distinct but related topics here. (1) proper response ordering across participating connections in a multi-connection session, for a handful of task/TMF responses (2) proper way to terminate and signal tasks when actions on one session can impact multiple initiators (1) is all about response fence, it is covered separately in section 3.3 of IG draft. That's what this email thread started off with. (2) covers cases that IG had called as as "multi-task abort" cases that typically have a shared task set across sessions, or are the result of a Logical Unit Reset TMF. Section 4.1 of IG adresses (2). So how are (1) and (2) related? - Some cases covered under (2) overlap with (1), but some cases in (2) are for single connection sessions. E.g. LU Reset on a single connection session impacting several 3rd party sessions - Some cases under (1) overlap with (2), but some cases in (1) have nothing to do with other sessions. E.g. establishment of ACA on a non-shared task set - Section 4.1 (i.e. text for (2)) uses response fence (i.e. text for (1)) whenever multi-connection sessions are involved in multi-task abort situations. I hope that clarifies it. Feel free to ask if that is not clear. The net is that the response fence is the underpinning notion to describe the correct iSCSI behavior in a few cases and some of those cases are about task terminations across third-party sessions. Mallikarjun----- Original Message ----From: John Hufferd <jhufferd@xxxxxxxxxxx> To: Mallikarjun C. <cb_mallikarjun@xxxxxxxxx>; ips@xxxxxxxx Sent: Friday, January 5, 2007 1:01:06 PM Subject: RE: Response Fence Flag Mallikarjun,I must admit, that I have been a bit confused with the direction of the conversation here.Therefore, I went back and reviewed the charts from Dallas . And as I remembered, the initial focus was to address the issue of Multiple Connections per Session (MC/S) (as stated on chart 4 – “Non-issue for single-connection iSCSI sessions”). So I think I have missed the step where we have morphed the discussion into one dealing with multiple sessions. (I am not sure how that happened, or if I miss-read the charts from Dallas , or have not followed the discussion adequately.)If we are attempting to define two different issues, one with MC/S and one with Multiple Session from different Initiators, I think it would be useful to break down the conversation into Topic A – MC/S and Topic B Multiple Sessions. It is possible that one solution will addresses both, but I for one think I am hearing arguments that might be appropriate for Topic B, while I am thinking about its applicability to Topic A.Perhaps, you could address the issue as either being all about MC/S or explicitly state that it is intended to affect Multiple Sessions also, and then address the issues and solution for each separately. For example, I believe Robert was addressing the issue from a view of Multiple Sessions and if we only intended to address MC/S then I expect the response might be somewhat different.Anyway, if you could clear-up some of this, I think it would be useful (at least to me).. . . John L Hufferd Sr. Executive Director of Technology Brocade Communications Systems, Inc jhufferd@xxxxxxxxxxx <mailto:jhufferd@xxxxxxxxxxx> Office Phone: (408) 333-5244; eFAX: (408) 904-4688 Alt Office Phone: (408) 997-6136; Cell: (408) 627-9606------------------------------------------------------------------------ *From:* Mallikarjun C. [mailto:cb_mallikarjun@xxxxxxxxx] *Sent:* Friday, January 05, 2007 10:08 AM *To:* ips@xxxxxxxx *Subject:* Re: Response Fence FlagNot really. Current draft text is intentionally written to not have any dependencies on T10 dynamics. The point is that iSCSI needs such a notion for succinctly describing the proper iSCSI protocol actions in a few places - ACA, TMF, Persistent reserve/Abort to name a few. We certainly hope it will be approved by T10 and be a part of SAM-4 soon, but that isn't required per se for describing what iSCSI needs for its correct behavior.IPS WG has adopted what it needs in the past - staying ahead of T10 review/approval cycle if necessary. I_T nexus loss notification, iSCSI target/port naming, clearing effects are a few I recall.Mallikarjun----- Original Message ---- From: Eddy Quicksall <Quicksall_iSCSI@xxxxxxxxxxxxx>To: Robert Snively <rsnively@xxxxxxxxxxx>; "Elliott, Robert (Server Storage)" <Elliott@xxxxxx>; ips@xxxxxxxxSent: Friday, January 5, 2007 8:58:47 AM Subject: Re: Response Fence FlagFrom an earlier email I think that Response Fence is only a proposal in T10 (http://www.t10.org:80/doc06.htm <http://www.t10.org/doc06.htm>). If so shouldn't iSCSI wait a bit until this has been ratified?Eddy __________________________________________________ Do You Yahoo!? Tired of spam? Yahoo! Mail has the best spam protection around http://mail.yahoo.com ------------------------------------------------------------------------ _______________________________________________ Ips mailing list Ips@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ips
_______________________________________________ Ips mailing list Ips@xxxxxxxx https://www1.ietf.org/mailman/listinfo/ips