RE: Methods, Inheritance, Exceptions, etc. (was: Re: Methods in SMIng ?)

["Andrea Westerinen" <andreaw@cisco.com>]

Frank, Your definitions helped, but I still feel that I am talking about
method signatures (execution on the agent side).  For example, a method to
Reset an interface's hardware needs to execute on the hardware and not on
the manager.  But, maybe this is about how the MIB attributes are
manipulated and not about the final location to execute the code?

For another example, look at RFC2925 that defines MIB entries to do pings,
traceroutes and lookups.  Pings are done at the remote host, at the request
of the management application, using either ICMP, the UDP echo port, timing
of an SNMP query, ...  "A single SNMP PDU can be used to create and start a
remote ping test."  Would you call this a method or procedure?

Andrea

-----Original Message-----
From: owner-sming@ops.ietf.org [mailto:owner-sming@ops.ietf.org]On
Behalf Of Frank Strauss
Sent: Thursday, May 03, 2001 12:38 AM
To: Andrea Westerinen
Cc: Juergen Schoenwaelder; david.putzolu@intel.com; sming@ops.ietf.org
Subject: Re: Methods, Inheritance, Exceptions, etc. (was: Re: Methods in
SMIng ?)


Hi!

Andrea> Totally disagree on your "disagree".  People already DO "do
Andrea> what the protocol does not allow."  You do have methods today
Andrea> - they are just not obvious.  All that I would like to do is:
Andrea> 1) Make them obvious in the definition/semantics, even if not
Andrea> obvious in the protocol

I guess we are talking about `procedures' here. Let me try to give
definitions of `procedure' and `method':

A procedure is a more or less formally defined algorithm that uses
protocol operations on MIB objects to achieve a specific high-level
goal. The interface of a procedure may be formally defined by a
signature. The algorithm is executed on the manager side.

A method is a specific high-level operation on an object instance of a
class. Its interface is formally defined by a signature. The operation
is executed on the agent side where the object lives.

Andrea>                           2) If EOS or another next gen protocol
Andrea> evolves to support methods, then you do have a protocol that
Andrea> handles the semantics more cleanly.

Maybe, some time we will have a protocol that supports method
invocations. I don't say that we will never want to have methods in
SMIng. Note that issue #52 claims for language extensibility and I
support this requirement.

As far as I know, current EOS efforts are concerned with some new
specific protocol operations, but not with a new generic `invoke'
protocol operation.

Andrea> BTW, I am not arguing to introduce new constructs.  I am
Andrea> arguing to use the existing constructs of readable/writable
Andrea> attributes to map the semantics of methods.  We do have
Andrea> methods today in MIBs - for example, we request a ping and a
Andrea> traceroute in remote ops.

I know what you mean. But I would not call it methods.  So, I think we
are talking about the same thing, just using different terms.

 -frank