Re: iSCSI - Change Proposal X bit

["Julian Satran" <Julian_Satran@il.ibm.com>]

That is an interesting proposal. However since it won't work in face of an 
initiator that does not actively use the connection (your two prameters 
are per/session values determined by the initiator.

I read Mallikarjun's proposal to add a stricter requirement on activity 
and thus avoid cleaning NOPs - as "send some non-immediate request that 
requires a response on every connection at least once in 2^31-1 CmdSN  to 
include anything sent (no additional traffic).

I intended to something close to your rule at the target to avoid 
increasing the window into the "dangerous zone" while not requiring a 
synchronous barrier at the initiator.

Julo




sandeepj@research.bell-labs.com (Sandeep Joshi)
Sent by: owner-ips@ece.cmu.edu
27-10-01 03:08
Please respond to sandeepj

 
        To:     cbm@rose.hp.com, ips@ece.cmu.edu
        cc: 
        Subject:        Re: iSCSI - Change Proposal X bit

 

Mallikarjun,

This problem can be solved _without_ introducing
any additional I/O activity.   Notice that it 
occurs because we allow the maxCmdSN to increase 
unbounded in the presence of unacked statSNs.

The target can prevent this from happening.

Imagine some connection A at the target :

(1) Everytime the target sends a new statSN out, it 
    records the cmdSN for that statSN 
    (say Connection_A.cmdSN_of_last_statSN = 200)

(2) While growing maxCmdSN at target, ensure that
    (maxCmdSN - Connection_A.cmdSN_of_last_statSN < 2^31-1)

    This is stricter than current condition
    (maxCmdSN - expCmdSN) < 2^31-1

    If there are N connections, use least value of 
    cmdSN_of_last_statSN.

(3) All retried commands will always fall outside 
                 the window.  They will be on the "dark side".

(4) If the target cant grow maxCmdSN, it closes the
    offending connection, or sends out NOPs.

Works ..?

-Sandeep


> In helping others walk through your scenario here in HP,
> we all realized that the core solution to the problem
> is to mandate a complete execution of a numbered command
> (some I/O activity that is trackable) on each connection
> once at least every (2^32 -1) commands.  It may be a NOP,
> or any regular command.  So, here is my suggestion -
> 
>   An initiator MUST send at least one non-immediate command
>   on each of the connections participating in the session,
>   for every (2^31 -1) numbered commands.
> 
> Comments?
> --
> Mallikarjun