No Subject

[Dave Nagle <bassoon@yogi.ece.cmu.edu>]

Message-ID: <39662389.91F0D45@agilent.com>
Date: Fri, 07 Jul 2000 11:38:01 -0700
From: Matt Wakeley <matt_wakeley@agilent.com>
Reply-To: Matt Wakeley <matt_wakeley@agilent.com>
Organization: Agilent Technologies
X-Mailer: Mozilla 4.73 [en] (WinNT; U)
X-Accept-Language: en
MIME-Version: 1.0
To: ips@ece.cmu.edu, Pierre Labat <pierre_labat@hp.com>
Subject: Re: Recovery from a dropped TCP connexion
References: <200007071819.OAA10554@yogi.ece.cmu.edu>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit


> Sender: plabat@cup.hp.com
> Message-ID: <396616B2.84603BDD@hp.com>
> Date: Fri, 07 Jul 2000 10:43:14 -0700
> From: Pierre Labat <pierre_labat@hp.com>
> Organization: Hewlett Packard ATM-SISL
> X-Mailer: Mozilla 4.51 [en] (X11; I; Linux 2.2.5-15 i686)
> X-Accept-Language: en
> MIME-Version: 1.0
> To: ips@ece.cmu.edu
> Subject: Recovery from a dropped TCP connexion
> Content-Type: text/plain; charset=us-ascii
> Content-Transfer-Encoding: 7bit
>
> Hello,
>
> In the last teleconf it has been stated that :
>  -> in case of a droped TCP connexion, a command that was on the flight
>       must be replayed on a new TCP connexion going through the same
>       service delivery port. This, because SAM-2 requires that all that
> concern
>       one command must flow through the same delivery port.
>
> But, because the recovery is done at the command level, why not
> abort this command on the target (the target will abort/cancel locally
> all the commands on the flight for the dropped TCP connexion),
> and replay the command on another link (other service delivery port).

The command cannot be aborted and replayed because some devices (tapes for
example) cannot abort a command and then replay it.  That is the whole reason
FCP-2 was created for Fibre Channel.

>
> As the command will be totally restarted we don't need some intermediate
>
> command state that would need to be maintained per port.
> We can cleanup what was ongoing for this command on the initial port
> and restart drom scratch on the alternate port.
>
> It doesn't hurt SAM-2 because the replay of the command is a whole
> new command.
> Of course iSCSI must change the LUN in the headers to take into account
> the new service delivery port.

But iSCSI does not have visibility to LUN mapping.  That is at a layer above
iSCSI.

>
>
> The big advantage is that it can speed up the recovery.
> 1) It is likely that if the TCP connexion dropped through a service
>   delivery port, the attempt to open a new one through the same
>   SDP will fail.
>
> 2) Replaying the command through an alternate link is faster
>   than waiting for the establishment of new TCP connexion.
>
> Regards,
>
> Pierre

Matt Wakeley
Agilent Technologies