Re: SNMP traps

["Thomas McSweeney" <rf42tpme@us.ibm.com>]

Mark,

I'm new to the iSCSI MIB effort, but I'll toss out some suggestions...

I think we should trap iSCSI Session terminations, except those
accomplished via Logout.  I know this might result in traps for "normal"
events, like when someone powers off their initiator instead of shutting
down gracefully.  If we want to get fancy, we could send them
conditionally, e.g., only trap if there were any incomplete commands or
undelivered responses at the time the session terminated.

Trapping iSCSI connection terminations would be interesting too, especially
to implementations which allow multiple connections per session.  Loss of a
connection could mean loss of capacity or reduced reliability of a session,
which could be reflected on the management station display.  This may
already be covered by the TCP MIB (I haven't checked yet), but coordinating
a TCP trap with iSCSI objects would be more difficult than having an iSCSI
trap.

It would also be helpful to trap changes to a target's LU list, and/or LUN
mappings.  Minimally, a simple trap to notify the manager that a change
occurred would be enough to prompt the management station to do the GETs to
refresh its tables.

Tom McSweeney
iSCSI Development, IBM Storage Networking
Email: rf42tpme@us.ibm.com
Phone: (USA) 919-254-5634  (tie line: 444-5634)
Fax:   (USA) 919-254-0391  (tie line: 444-0391)


Mark Bakke <mbakke@cisco.com>@ece.cmu.edu on 03/28/2001 09:57:47 AM

Sent by:  owner-ips@ece.cmu.edu


To:   Sandeep Joshi <sandeepj@research.bell-labs.com>
cc:   ips@ece.cmu.edu
Subject:  Re: SNMP traps



Sandeep-

We have been looking for requirements for traps, and have not
addressed them yet.  You are the first to really ask for them.

Trapping the login and authentication failures seems like a
good idea; do you have any other ideas or requirements?

Thanks,

Mark