>As many of you have discovered, the SNMP protocol does not
>accomodate multiple agents in one device very elegantly. This
>has been a known problem for at least four years, with multiple
>solutions proposed during that time.
>
>Unfortunately, this problem does affect the design and
>implementation of the printer MIB on some printers, and
>constrains the jobmib work, as well.
I appreciate the pointer to the SubAgent group, however, I'm not
sure multiple agents per device is as prevalent a problem with
printers as is the device with multiple interfaces.
>This IETF Working Group, printmib, is *NOT* chartered to do this work.
>If this is an issue for your current and future product
>implementations, then I strongly recommend participating or monitoring
>the work in the agentx working group.
I don't think the PWG wants to engage this charter but we do want
to understand and attempt to adhere to the guidance we received from
the IETF during development of RFC1759. It seemed quite clear, then,
that a printer should have one agent and the Interface table of MIB-II
should be used to describe the interfaces (if not, why does RFC1759
specifically call out the Interface table of RFC1759?).
(I would draw a stick figure here, but I don't want to get Don all
riled up;-)
We did also receive guidance on the multiple agent problem - but this
only pertained to a specific scenario where a single (external) network
attachment services more than 1 printer (a scenario we probably wish
we had never mentioned and one that has never shown to be very
applicable to the network management scene in the long run. As I recall,
we used the term "Agent Orange" back then, which appears to have
evolved into the kinder, gentler AgentX, and our solution for this was
to index every SNMP response by the hrDeviceID of the printer it
belonged to.
It's been a pain, but, kept at this level, I think it works.
When we try to get too granular with this scheme, then we run into
problems. Gail has demonstrated this lately with each device (printer)
sharing some other device (storage) in the hrMIB. But, I don't think
this is the "multiple agent problem" (is it?). This is lack of
demarcation of shared resources (and I haven't quite decided how to
respond to Gail's dilemma).
I don't know what prompted your note on AgentX. I admit, I should
read up more on it. But, if you're saying we shouldn't try to
follow the advice given to us regarding Interfaces and hrDeviceID,
I'm not sure I agree.
Harry Lewis.