Re: iSCSI delay for session shutdown

[julian_satran@il.ibm.com]

Pierre,

Initiators do recovery. Target are supposed to wait and being told what to
do.
The connection recovery will be slightly changed and there will some
timeouts (I've mentioned
earlier on the bridging discussion) that we are going to use.

Julo

Pierre Labat <pierre_labat@hp.com> on 17/10/2000 01:26:50

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

To:   ips@ece.cmu.edu
cc:
Subject:  iSCSI delay for session shutdown




Julian,

Consider the following case:
- an host  system has two NICs (NIC1 and NIC2)
- the target has only one NIC.
- the host (initiator) establishes one session through
    one of its NIC  (NIC1) to the target.

After a while, NIC1 fails.
The host, to recover, decides to use NIC2, it opens a new
TCP connection (same session) through NIC2 to the target.

My question are:

1) how much time a target will keep a session "valid"
from the time the TCP(s) connection(s) can't pass
any traffic  (because
on the other side (initiator)  it is broken)?

It is no sure that it will be the time till TCP gives up depending
on the target implementation.

If this time is too short, when the initiator will open
a new TCP connection to recover, the target would
have dropped the session and the recovery will not work.

2) Could it be possible to specify a minimum delay before
the target drops the session?
Only in the case where the target doesn't receive
a logout notification of course.

This minimum delays would secure the initiator recovery
mechanism.

Regards,

Pierre