|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: iSCSI: Sequence Reception TimeOut for Response PDUs
To clarify the intent of 6.12.1 we may want to make the text of clause b) at the initiator:
Sequence reception timeout (no status) or response reception timeout - dealt with as specified in Section 6.6 Sequence Errors, using the option of a SNACK.
The added words are "or response reception timeout".
Julo
----- Forwarded by Julian Satran/Haifa/IBM on 06/30/2002 04:03 PM -----
| Julian Satran
06/29/2002 04:38 PM
|
To: "Praveen Kumar" <praveen@basilnet.com>
cc: ips@ece.cmu.edu
From: Julian Satran/Haifa/IBM@IBMIL
Subject: Re: iSCSI: Sequence Reception TimeOut for Response PDUs
|
Praveen,
Looking at the text again before putting it I think It would be better we leave things as they are and here are the reasons:
- Outstanding statuses/responses (i.e., those about which both initiator and target know) is handled in 6.12.1 and it includes timeout at both initiator and target
- The above leaves for 6.12.2 mainly those cases that are not known to both parties (one way messages) and those are best handled by the party that knows about them - and is covered by 6.12.2 target action on unacked statuses. The initiator may use a periodic NOP (as we say somewhere else) but the target should be left to worry about statuses about which the initiator does not know about.
So my latest suggestion is - LEAVE THINGS AS THEY ARE.
Sorry for the confusion I may have created.
Julo
----- Forwarded by Julian Satran/Haifa/IBM on 06/29/2002 04:24 PM -----
| Julian Satran
06/29/2002 12:27 PM
|
To: "Praveen Kumar" <praveen@basilnet.com>
cc: ips@ece.cmu.edu
From: Julian Satran/Haifa/IBM@IBMIL
Subject: Re: iSCSI: Sequence Reception TimeOut for Response PDUsLink
|
Praveen,
I intend to add to 6.12.2 the following text:
- An initiator may also recognize a lost iSCSI numbered response by an iSCSI timeout. In this case, error handling is done as specified in Section 6.6 Sequence Errors, using the option of a SNACK. However since SCSI command response is highly command dependent the SNACK may result in a Reject with an "invalid PDU field" reason code if the response was not issued.
That should solve the issue.
Julo
| "Praveen Kumar" <praveen@basilnet.com>
06/26/2002 10:05 PM
Please respond to "Praveen Kumar"
|
To: Julian Satran/Haifa/IBM@IBMIL
cc:
Subject: iSCSI: Sequence Reception TimeOut for Response PDUs
|
Julian,
In Section 6.12.2, under the Paragraph "-Lost iSCSI Numbered Response" , is there any reason for NOT including "Sequence Reception TimeOut" as one of the mechanisms for detecting a "Lost iSCSI Numbered Response"?
If a Numbered iSCSI response PDU is dropped (owing to header digest errors) , the Initiator could TimeOut waiting for the appropriate StatSN (assuming that the target implementation does Not support the issue of Nop-In, when Status is Not acknowledged for a long time).
Section 6.12.1 does mention this, but that section only deals with Lost Data or R2T PDUs, NOT Lost Response PDUs.
I would have expected this section (6.12.2) to mention this (Sequence Reception TimeOut) as a possibility (especially since the other sections explicitly mention this).
Thanks in Advance
Praveen
Home
Last updated: Mon Jul 01 01:18:49 2002
11026 messages in chronological order
|