Re: iSCSI: error recovery

[julian_satran@il.ibm.com]

Pierre,

Nevertheless please look for holes.  It will help us all.

Thanks,
Julo

Pierre Labat <pierre_labat@hp.com> on 09/11/2000 19:27:35

Please respond to Pierre Labat <pierre_labat@hp.com>

To:   Julian Satran/Haifa/IBM@IBMIL
cc:
Subject:  Re: iSCSI: error recovery




julian_satran@il.ibm.com wrote:

> Pierre,
>
> Interesting scenario - but ENTIRELY WRONG.
> A more carefull reading of the draft would have solved your problem.
> After a failed connection the two parties (I & T) are supposed to do some
> cleanup.
> In the old draft that was accomplished by having the initiator indicate
in
> the new login
> what old connection it is replacing.

In the old draft, to do the cleanup, opening a new connection was
mandatory.
Of course opening this new connection has to be quick because
the recovery is waiting for that to start.
But opening a new connection because one failed is not always
easy and fast, you can easily hit  the same problem that failed
the original one.
Hence i was thinking (among other options) about recovering
with the already opened connections (without the cleanup that
would have been done by opening a new connection).
The scenarios can apply in this case.

But now that we have the logout message that allows to do the cleanup
without opening a new connection (as far as you have at least two
connections in a session), the scenarios i exposed are solved.
Hence if the session has several TCP connections
and one breaks, you can start the recovery right away on an already
opened connection and
try to re-open the failed TCP connection in background.

Regards,

Pierre