RE: Comments from the meeting on Monday and the draft

["Philip J. Nesser II" <phil@Nesser.COM>]

Michael,

As much as I appreciate the elegant concepts in the whole OID model, I have to say that it is not likely to be apreciated as a solution to this problem.  One of the reasons that I heard from community pushing for this document is the failure of SNMP to solve these types of issues.  Subsuming an integral part of SNMP as a part of the solution space here is not likely to be popular.

As for a lifecycle I would suggest that either a group be spun off to spend time thinking about a structure and mandate a series of required codes.  It is also possible to make the number space large enough to be reasonable for the forseeable future.  (i.e. use 5, 6, 7, etc... digits) to allow millions of possible codes.  The goal is to allow people developing code to implement configuration management programatically.  Writing a parser to deal with 7 digit codes is not much harder than 4 digits.  

--->  Phil  

-----Original Message-----
From: owner-ops-nm@ops.ietf.org [mailto:owner-ops-nm@ops.ietf.org]On
Behalf Of Michael MacFaden
Sent: Wednesday, December 12, 2001 5:58 PM
To: Philip J. Nesser II
Cc: ops-nm@ops.ietf.org
Subject: Re: Comments from the meeting on Monday and the draft


On Wed, Dec 12, 2001 at 01:32:43PM -0700, Philip J. Nesser II wrote:
>Sounds reasonable.  But if we want to leave it in (and just to be clear I think its a reasonable idea), then we at least need to define the basics.  ie 
>
>Response codes are 4 digit decimal values.  The first digits specify a catagory or classification of response type.
>
>The following catagories are defined:
>
>1XXX:  Sucessful Completion
>2XXX:  Syntax Errors
>3XXX:  Security Problems
>4XXX:  etc...
>5XXX:  etc...
>6XXX:  etc...
>7XXX:  etc...
>8XXX:  Vendor Specific Responses
>9XXX:  Reserved

>(Clearly more thought needs to go into this than I have just tossed off.)

Once you go down this path, you end up needing a lifecycle model 
as well. Can you reuse numbers without regard to some forward
and backward compatiblity rules? Can the semantics be well defined such
that any program can reasonably do anything with the categories
defined above? 

Seems to me OID concept and OBJECT-IDENTITY macros (RFC 2578/STD 58) 
from the SMI already solve the problem of binding a semantic to
a globally unique identifier a long time ago in a flexible and 
scaleable way....  even though OIDs are only slightly less ugly to 
humanly parse than IPv6 addresses.

There was an old joke I recall that went something like this:

...The fall of the Roman Empire was due to the fact they could not
return 0 to indicate success from a subroutine in their C programs....

Regards,
Mike MacFaden