Re: Syslog-sign: Last minor clarifications/nits

On Mar 24, 2009, at 2:13 PM, David Harrington wrote:

> Hi,
>
> why are we allowing multiple ways to set this?
> Since our mission is to define standards, why not standardize one way?
>
> I think there are problems with rollover when using unix time or
> snmpBoots.
> If we simplify this to use a simple increasing value, then we avoid
> that complexity.

You are correct, but there will be the people who will fold in the  
other ways as well. If that's okay, that's okay.

There's a problem that simplicity is in the eye of the beholder.

A mere counter is perhaps the simplest, but could lead to operational  
issues, especially if you want to tie it together.

A timestamp is simple in the sense that it's easy to get, but has  
rollover issues. Those are easy to deal with (including by making it  
be a 48-bit base64 value as Kelsey first had it), but you have to deal  
with them.

SNMP values are simple if you have an SNMP framework, but not simple  
if you don't.

What you said, "simple increasing value" simplifies definition, but  
pushes issues off to implementation and operations folks, who will  
pick one of the other obvious things -- they're each simple increasing  
values. Timestamp is merely sparse.

Arguably even simpler is Kelsey's definition which said it was non- 
repeating, as opposed to increasing.

But are they really any different, at the end of the day? The major  
debate I can see is whether it should stay a ten-digit decimal number  
or go back to a base64 48-bit binary number.

	Jon

_______________________________________________
Syslog mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/syslog