Re: a vote for asymmetric connections in a session

["Matt Wakeley" <matt_wakeley@agilent.com>]

Joshua Tseng wrote:

> >As Julian has stated in a different thread, the purpose of the "sliding
> >windows" in iSCSI is not for congestion management.  It is simply there to
> >handle the case where if a connection goes down in a multiple connection
> >session, it prevents the remaining connections from overwhelming the target
> >with new commands that it can't process due to missing commands that where
> on
> >the broken connection.
>
> Matt,
>
> In the event of a failed connection, the sliding window isn't going to help
> much since you have to first detect the failed connection.  Once this is
> accomplished, the initiator will know which command was lost for sure,

How is the initiator going to know what commands got to the target or not
without the command reference numbers in the iSCSI document?

> but will not know how many, if any, of the subsequent commands sent after the
> lost command, were also lost.

Well, all of them on the failed connection - if the connection failed, the
following commands could not have been delivered.

>  I presume, the initiator will then
> retransmit all commands without knowing whether or not they were lost,
> rather than wait for the target to advance the acknowledgement counter.
>
> This is all awesome and impressive behavior, exactly what we want.  The only
> problem is this exact same behavior can be had without the windowing
> mechanism.

Could you describe how it can be had?

>
>
> >Since all of this runs on top of TCP, and TCP performs congestion
> >management, why must iSCSI perform congestion management on top of TCP?
>
> Please see my reply to John Hufferd.
>
> Josh

-Matt