RE: rationale to send PDUs in increasing CmdSn on single co nnection

Agree with Julian on this.

RFC 3720 simply requires that once the CmdSN is
assigned to a command, it has to go out in that order
on to the wire.  Note that RFC 3720 does not mandate 
where/when CmdSN assignment must be done, other than
of course that it is done before the command goes out.
 I don't see the specific DMA optimizations that the
RFC precludes.

Mallikarjun 


--- Julian Satran <[email protected]> wrote:

> As assigning CMD sequence number is the prerogative
> of the initiator I do 
> not  understand the point of send out of order
> commands. If you want to 
> optimize within an independent RDMA layer at the
> initiator the assign the 
> numbers there.  Initiators sending out-of-order
> packets will probably be 
> flagged as bad by any of the certification packages.
> 
> Julo
> 
> 
> 
> Parav Pandit <[email protected]> 
> 01/09/07 09:18
> 
> To
> David Sheehy <[email protected]>, Julian
> Satran/Haifa/IBM@IBMIL
> cc
> [email protected]
> Subject
> RE: [Ips] rationale to send PDUs in increasing CmdSn
> on single co nnection
> 
> 
> 
> 
> 
> 
> Exactly thats the reason which lead me to asko this
> question.
> 
> DMA optimization is possible if we can send out of
> order commands to the TCP layer on single
> connection.
> 
> If I take pragmatic approach, 
> Any idea, currently how current single connection
> target implementation is?
> 
> I mean, if current targets are smart enough (accept
> out of order PDUs on single connection- liberal on
> point 3.2.2.1), then initiators can use their DMA
> optimization.
> What do you say?
> 
> Parav
> 
> 
> --- David Sheehy <[email protected]>
> wrote:
> 
> > > and there was no visible motivation for out of
> > order commands on a single connection 
> > 
> > On the contrary, there was plenty of motivation.
> It
> > was overridden by the desire for simplified
> > implementations over single TCP connections as the
> > Parav conjectured. There are DMA related
> > efficiencies that can be realized in situations
> with
> > a mix of solicited and immediate/unsolicited I/O
> > traffic. The folks in favor of this optimization
> > were outvoted by those who wanted to simplify
> their
> > single connection based implementations.
> > 
> > Dave
> > 
> >   _____ 
> > 
> > From: Julian Satran
> > [mailto:[email protected]] 
> > Sent: Friday, August 31, 2007 8:54 AM
> > To: Parav Pandit
> > Cc: [email protected]
> > Subject: Re: [Ips] rationale to send PDUs in
> > increasing CmdSn on single connection
> > 
> > 
> > 
> > 
> > Parav Pandit <[email protected]> wrote on
> > 31/08/2007 12:52:04:
> > 
> > > Hi,
> > > 
> > > RFC 3720, section 3.2.2.1 says 
> > > 
> > > "On any connection, the iSCSI initiator MUST
> send
> > the
> > > commands in increasing order of CmdSN, except
> for
> > > commands that are retransmitted due to digest
> > error
> > > recovery and connection recovery. "
> > > 
> > > (Assuming Single TCP connection ISCSI session)
> > > 
> > > 1. I interpret above 3.2.2.1 statement as 
> > > SCSI layer gives SCSI commands to the ISCSI
> stack
> > in
> > > the order of Cmd-1 and Cmd-2. 
> > > Cmd-1 will have CmdSn = 10.
> > > Cmd-2 will have CmdSn = 11.
> > > ISCSI stack CAN send PDUs to the TCP layer in
> > > following order ONLY.
> > > PDU-1 with Cmd-1.
> > > PDU-2 with Cmd-2.
> > > 
> > > Is this correct interpretation?
> > > Or
> > > 
> > Yes 
> > > 2. On a SINGLE connection can ISCSI stack send
> the
> > 
> > > PDU-1 with Cmd-2 followed by 
> > > PDU-2 with Cmd-1?
> > > 
> > NO
> > > Assuming the answer of the question #2 is No,
> > > 
> > > 3. If there are multiple connections in a
> session
> > then
> > > command MAY any way reach out of order. And
> > targets
> > > need to wait for the previous expected commands.
> > > 
> > > So targets will receive out of order ISCSI PDUs
> > from
> > > the TCP layer and ISCSI stack handles them.
> > > 
> > > So then why initiators have restriction of
> sending
> > > command in the increasing order of CmdSn on
> SINGLE
> > TCP
> > > connection?
> > > 
> > To simplify recovery and to...
> > > Is it to simplify the implementation of targets
> > > supporting only single TCP connection?
> > > 
> > > 
> > 
> > and there was no visible motivation for out of
> order
> > commands on a single connection 
> > 
> > > Regards,
> > > Parav Pandit
> > > 
> > > 
> > > 
> > > 
> > >
> >
>
____________________________________________________________________________________
> > > Looking for a deal? Find great prices on flights
> > and hotels with 
> > > Yahoo! FareChase.
> > >  <http://farechase.yahoo.com/>
> > http://farechase.yahoo.com/
> > > 
> > > 
> > > _______________________________________________
> > > Ips mailing list
> > > [email protected]
> > >  <https://www1.ietf.org/mailman/listinfo/ips>
> > https://www1.ietf.org/mailman/listinfo/ips
> > 
> > 
> 
> 
> 
>  
>
____________________________________________________________________________________
> Luggage? GPS? Comic books? 
> Check out fitting gifts for grads at Yahoo! Search
>
http://search.yahoo.com/search?fr=oni_on_mail&p=graduation+gifts&cs=bz
> 
> 
> _______________________________________________
> Ips mailing list
> [email protected]
> https://www1.ietf.org/mailman/listinfo/ips
> 
> > _______________________________________________
> Ips mailing list
> [email protected]
> https://www1.ietf.org/mailman/listinfo/ips
> 



       
____________________________________________________________________________________
Building a website is a piece of cake. Yahoo! Small Business gives you all the tools to get online.
http://smallbusiness.yahoo.com/webhosting 


_______________________________________________
Ips mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ips