Editor's note: These minutes have not been edited. Minutes of Entity MIB Working Group LA IETF - 4,7 March 1996 Chaired by Keith McCloghrie Minutes by Bob Stewart with assistance from Andy Bierman 4 March Along with his nice leather Cisco notebook, the minutes taker lost the minutes for the first day when he left the notebook on a chair in the meeting room after the Open IAB Meeting. Maria Greene presented a proposal for removing the community string object in favor of a more abstract object that represented a naming scope without giving away a full community string. She included a proposal for formatting community strings using that object. We discussed this for most of the meeting and concluded that we didn't agree and wouldn't change the MIB. Maria presented a definition of a logical entity which we liked. I think Maria presented something else, too. 5 December Keith suggested a consensus that the Entity MIB is allowed but not required unless it has something to point to. Discussion ensued: - Can there be a hierarchy of different contents? - We want one place with all the information. What we have now is that. - That's a Chassis MIB and goes too far for our constraints. - We're trying to avoid consistency problems. That's a general problem we can't solve. - The problem arises with multiple sources and attempting to be flat. - Relationships are beyond the scope. We only need a directory of logical entities. - Must we preclude different transport addresses for different logical entities? - We should stay simplest possible. - How do we account for multiple paths to the same information? - Why do we need a transport address? - Some agents may multiplex logical entities by transport address rather than community string. - Consensus reached on Keith's suggestion with no document changes. Maria Greene presented a proposed Johnson-Greene index: - Maximum hierarchy depth of four using two indexes. - How about level/parent/index? - Let's drop hardware. We're spending too much time on it, and it's not our primary need. We should finish logical entities and do hardware separately. - A subgroup will go aside to try to resolve problems with this. - So far this proposal doesn't help. - Support dropping hardware. - The subgroup reported failure to resolve problems. - We reached no consensus for changes. We discussed issues list from Dallas: - SNMPv2 is removed. - We have sufficiently discussed the MIB structure. - We decided that after a reboot the hardware must get the same indexes unless the hardware configuration is different. . How do you tell if it's changed? Sometimes you can't. . This is a useless feature. Throw out hardware. . Add an object to count re-indexings. . Solutions exist but are difficult and use NVR. . We need an unchanging identification. . We alread have OID, descriptor and others. - We have sufficiently discussed containers and containees. - We haven't found a sufficient standard list of physical class values. . Use the list from the Host Resources MIB. . It has no overlap with our list. . Our lack of trivial expandibility is a feature. . Remove power supply and fan. . No. . We'll leave the list as it is in the current MIB draft. - We have added and deleted objects as needed. - For conformance we will allow logical only. - We have covered Trap issues. - We decided an agent may not return only a subset of logical entities. - The definition of a logical entity is as presented by Maria on the 4th. - sysORTable issues: . It's a list of MIBs available. . Does an entity with limited scope list only its MIBs? . Yes. . In any case, you can't count on it being exactly correct. - We accepted Maria's physical containment table. Andy presented some discussion on the entAliasMappingTable: - It's currently too complex. Changes proposed. - Changes accepted: switch to physical then logical index and use OID for port, remove entAliasMappingIndex as index. We need definitions relative to hosts for backplane, chassis, etc. Text will be added, but not too definitive. We discussed per-class numbering: - Separate number ranges are good for separate classes. - Modeling slots moves the parent around. - You should number within a container and look up a level for relative position. - No change. We have nothing more to say about multiple agents. The editor will have an updated document next week. We discussed adequacy of compliance and interoperability testing: - entLogicalType and descriptor strings need improvement. - We should add an OID for descriptor. - Descriptor is already redundant with the existing OID. - The intent was to display the descriptor and use the OID for interoperability. - Further discussion on this topic will be on the mailing list. Respectfully submitted, Bob Stewart