Re: Referential integrity across reboots

[Andy Bierman <abierman@cisco.com>]

At 08:22 AM 8/5/2002 -0700, David T. Perkins wrote:
>HI,
>
>With experience has come wisdom in MIB design, which is that 
>resource instances that persist across reboots MUST have names
>(instance identifiers) that also persist across reboots.
>
>IF entries are examples of persistent resource instances.
>Processes are examples of non-persistent resource instances.
>
>The original design of the IF table had some problems.
>The graduate level question for MIB designers is for persistent
>index values, what should the approach taken so that their
>values are persistent across reboots? One cannot assume that
>all managed devices have "enough" non-volatile storage to save
>all persistent values. Also, one cant not assume that StorageType
>objects are fully supported. (That is, in most devices StorageType
>cannot be supported because saving instances of configuration
>values is not supported. Only the complete configuration can
>be stored.)
>
>Hope this helps....

yes -- I agree 100%.  Experience has demonstrated that the ifIndex
and entPhysicalIndex type of MIB design is flawed. (This is yet another
reason why CLI much more useful than MIBs for configuration.)  It is
critical that persistent resources retain the same identity across
reboots.  I guess the IF-MIB, Entity MIB, and any other similarly 
flawed MIB should be fixed.

Andy



>At 10:33 AM 8/5/2002 -0400, Mike Daniele wrote:
>>Andy Bierman wrote:
>> 
>>> The solution is to simply mandate that instances preserved
>>> across a reboot come back with the same values.  Currently,
>>> it's a SHOULD, not a MUST.
>>
>>Hello,
>>
>>What exactly is it that's currently defined as SHOULD?
>>
>>In general, it sounds like a great idea to make index values
>>persist across reboots.  But I'll make 2 obvious observations:
>>
>>1. It's not possible for all index values (for example, a process ID).
>>
>>2. ifIndex is clearly defined as being permitted to change across
>>reboots.  So are you suggesting a new index clause for ifEntry?
>>
>>Mike 
>Regards,
>/david t. perkins