|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: iSCSI question
Amir,
Which
one requires more CPU processing?
-Shahram
Shahram,
1)
Connection reassignment requires more than PDU retransmission to
handle corner cases
2) Error recovery level 1&2 requires more buffering which
takes space and lowers performance. If
you support error recovery level 1 you might as well
support level 2.
Amir
Pat,
Thanks. I understand your point. Although
terminating a session may be easy, but, starting a new session requires
new login, parameter exchange, new connections establishment,
authentication, etc. So I wonder how is this any simpler than a simple PDU
retransmit?
Yours,
-Shahram
Shahram,
Wen you start
a new session, you don't recover any PDUs. All the iSCSI state died with
the old session. iSCSI doesn't know the new session had any relationship
to the old session.
As Julian
said, recovery at that point is up to the SCSI layer above iSCSI. It is up
to SCSI to retry any commands that it wants to retry. When SCSI retries a
command, iSCSI doesn't know it is a retry. To the iSCSI layer it is just
like any other SCSI command it receives.
Pat
Julian,
To start a new session you need to start new
connections and you need to support
the PDU recovery. So how is that a subset of PDU
and connection recovery?
-Shahram
(I will explain the detailed clarity issues in
another email)
Session
recovery is in fact leaving all recovery to SCSI - it drops everything
and creates a new session. As
for you comment on the clarity of chapter 5 at this stage it makes sense
to be either specific or keep
this type of comment out of this context.
Julo
| Shahram Davari
<Shahram_Davari@pmc-sierra.com>
08/07/2002 06:09 PM
|
To: Julian
Satran/Haifa/IBM@IBMIL cc:
ips@ece.cmu.edu, owner-ips@ece.cmu.edu Subject:
RE: iSCSI question
|
Julian, Thanks. I
have read that section but it is not very clear. I also agree that Connection recovery requires
everything in command recovery. But what about session recovery? isn't it a superset of both
connection and command recovery? Yours, -Shahram -----Original
Message----- From: Julian Satran
[mailto:Julian_Satran@il.ibm.com] Sent: Wednesday, August 07,
2002 11:03 AM To: Shahram Davari Cc:
ips@ece.cmu.edu; owner-ips@ece.cmu.edu Subject: Re: iSCSI
question
Sharam,
You may want to go over the
recovery chapter.
It has detailed answers to all
your questions. The superset/subset is based on functions you
need for the next level.
Session recovery drops real
recovery to SCSI.
Command recovery recovers from
individual command errors without changing connection and the
highest enable you to switch to a new connection and continue commands there.
2 requires
everything in 1.
Julo
| Shahram Davari
<Shahram_Davari@pmc-sierra.com> Sent by: owner-ips@ece.cmu.edu
08/07/2002 05:17 PM
|
To:
ips@ece.cmu.edu
cc: Subject:
iSCSI question
|
Hi,
I have
a question regarding the hierarchy of error recovery. Section 6.13
mentions the hierarchy as:
2: Connection recovery 1: Digest
failure recovery 0: Session recovery
And it states that the
higher levels are a superset of the lower levels and that the level
of complexity increases from 0->1->2.
Couple of
questions:
1) How is digest failure recovery done? by
retransmission of PDUs? 2) Why is the connection recovery a superset
of session recovery and more complex? 3) It seems to me the order
should be:
2: Session recovery 1: Connection recovery 0:
Digest failure recovery
I appreciate any
insight.
Thanks, -Shahram
Home
Last updated: Thu Aug 08 17:18:58 2002
11576 messages in chronological order
|