RE: same value operation attributes restriction

[Andy Bierman <abierman@cisco.com>]

At 02:55 PM 3/26/2004, Wijnen, Bert (Bert) wrote:
>Inline
>
>> -----Original Message-----
>> From: Andy Bierman [mailto:abierman@cisco.com]
>> Sent: vrijdag 26 maart 2004 23:42
>> To: Wijnen, Bert (Bert)
>> Cc: Gilbert Gagnon; Glenn Waters; j.schoenwaelder@iu-bremen.de;
>> netconf@ops.ietf.org
>> Subject: RE: same value operation attributes restriction
>> 
>> 
>> At 02:28 PM 3/26/2004, Wijnen, Bert (Bert) wrote:
>> >Andy, w.r.t. your examples below, you explain (case B) how 
>> >difficult it is on the manager site if 3 PDUs need to be used.
>> >But if you use case A, does that complexity then not need to
>> >be dealth with at the managed device if for exampel an error
>> >occurs during the <boo operation="modify"> .... </boo> phase?
>> 
>> Not if rollback-on-error is used.
>> The agent will save the start state and restore it
>> if any of the 3 operations fails.  If ignore-error
>> or stop-on-error is used, then the manager has to
>> closely examine the <rpc-error> response and take
>> action accordingly -- this would be the same effort
>> whether approach A or B is used.
>> 
>> 
>> >And if so, then why do you think/expect that that is easier on 
>> >the managed device?
>> 
>> It's not easier on the managed device.  Rollback-on-error
>> is non-trivial to implement (that's why it's a capability).
>> 
>So...
>- It IS important to make that VERY CLEAR to everyone.
>  One of the complaints of SNMP was that WRITE and CREATE support
>  was too difficult to implement correctly in an agent.
>  So do we evaluate if this is "doable" in a managed device!?

Many customers are asking for it.
Some vendors are working on implementing it, whether NETCONF
has it or not.

>- You say it is a capability. So managers may not assume that the
>  facility is available and so in cases where it is not supported
>  they would still have to do the complex recovery/correction, no?

yes -- but the error recovery complexity is not coupled
to the number of different operation attribute values.
Complex 1 operation PDUs will be harder to undo than
simpler N operation PDUs. (I.e., change all the attributes
to the same value in my example and it doesn't affect
the complexity to undo foo, goo, and boo at all.)



>> One of the significant reasons that SNMP failed for configuration 
>> is the assumption that agents should be dumb and all the transaction
>> complexity should fall on the manager.  IMO, the agent needs to 
>> provide appropriate transaction capability, even if that's not easy.
>> 
>
>Sounds as if it should be mandatory instead of a capability ?

IMO, it should be a capability because it is difficult to do.
Market pressure will dictate whether devices without this
capability will succeed.


>Bert 

Andy




--
to unsubscribe send a message to netconf-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.
archive: <http://ops.ietf.org/lists/netconf/>