Re: Mib questions

[Bob Stewart <[e-mail]>]

Dave, those are really neat questions, simple and to the point.  I find I have
my own quick answer to most of them.  It will be interesting to see if I match
anybody else.  All and sundry, beware that these answers reflect my personal
understanding and prejudices regarding what we're doing.  They are NOT an
official interpretation.  If they generate shock, then Dave's questions were
particularly pertinent.  If they generate no disagreement, I'll be surprised.

>1) What is the scope of the chassis MIB?  Is it limited
>   the the devices covered in the Overview (section 5)?

It can include any abstract "networking device," including but not limited to
the specific examples in the overview.  It could, for example, include a host.

>2) Do the terms "functional module" and "logical device"
>   mean the same and are they things like routers, bridges,
>   terminal servers, etc?

I couldn't find "functional module."  "Module" seems to always mean physical
module.  Logical devices are things like routers, bridges, etc., and may or
may not match up one-to-one with physical modules.

>3) Is there a requirement that the "logical devices" have
>   network connectivity to each other (ie be able to pass
>   traffic to each other)?

No, assuming that "network connectivity" and "traffic" mean IP traffic or
equivalent. 

>4) Can any "logical device" implement the chassis MIB?

Yes.

>5) Must all the "logical devices" implement the chassis MIB?

No.

>6) If more than one implements the chassis MIB, must all the
>   values be "consistent"?

Muahahahahaha.  Yes, on the basis that consistent means different but related
as opposed to identical.

>7) What is the minimal requirements for gathering information
>   that a "logical device" must have when implementing the MIB?

It must have some communication path, perhaps out of (network) band, to obtain
the information.  It should do this either on request or periodically as long
as the information is timely (whatever timely means).

>8) Is it valid that all the information in the slot table (and
>   most other MIB objects) be entered by a "configuration file"
>   that an agent loads at its boot time?

I'd say that's technically valid, but could be severely misleading due to
timing of changes and update.  I'd expect competition to unmercifully bloody
such a product.

>9) What are some situations the it would be appropriate to use
>   a "sparse" slot table?

Any situation the implementor chooses.  You might choose to do that in a
2-slot chassis or one the size of Texas.

>10) Is a "logical device" a realised hardward component which has
>    an unchanging identity, or does it also include "personalities"
>    that are identified by the software that is running on the
>    "logical device". An example would be a File server, loadable
>    RMON software, Mail server, or any other "application server".

Logical device overlays one or more hardware components, and is thus
"personalities."  With appropriately fancy hardware and software, several
slots could conspire, for example, to implement a logical terminal server, a
logical router, AND multiple logical repeaters.  Or, more simply, a single
slot may contain a single terminal server.  Or, slightly more complex, a
single slot with multiple configurable ports may implement multiple logical
repeaters, one for each network it repeats.

>A fundamentation question is just how is an agent that implements
>the chassis MIB suppose to get all the information about the other
>"logical devices" - 1) through SNMP requests to the other "logical
>devices", 2) through a private network protocol, 3) through a backplane
>bus using shared memory or messages between CPUs,  or 4) through some
>"static" configuration file that is loaded at agent boot time (and
>maybe refreshed during running)?

Simplistically, the answer is completely implementation dependent, but I have
certain expectations.  I'd say number 1 would be at least semi-silly.  In my
mind the Chassis MIB implies a closer sense of chassis community than that,
and I see no reason to believe all the entities in a multi-segment chassis
will have normal network connectivity.  Number 2 is about the same as number
1, a bit better or worse depending on the origin of the private protocol.
Number 3 is what I envision as most sensible.  Number 4 could be viewed as
abstractly the same as number 3, but has problems of timelyness.

	Bob