Add note about LLDP brokenness

Author: ytti

LLDP-IS-BROKEN.md

@@ -0,0 +1,71 @@
+Someone with energy to work on standard bodies should drive LLDP change in such
+way that it is actually consumable by vendor-agnostic code.
+Today working LLDP code would require per-vendor models to translate portID and
+chassisID into something consumable, often cases requiring heuristics.
+LLDP is extremely low value and with minor changes could be actually standard
+tool to discover network.
+
+
+```email
+[802.1 - 9431] LLDP portID
+To: [email protected]
+Subject: [802.1 - 9431] LLDP portID
+From: Saku Ytti <[email protected]>
+Date: Wed, 9 Jan 2013 15:03:08 +0200
+Delivered-to: [email protected]
+Reply-to: Saku Ytti <[email protected]>
+"Insanity is doing the same thing and expecting a different result."
+Before sending a duplicate post, see "Sending->retrying" at
+802.1 list help: www.ieee802.org/1/email-pages/zuwz1011.html
+-----
+
+Hi,
+
+Why is there no snmp ifIndex subtype for portID?
+
+It seems like obvious choice, which vendors actually do use. Only
+problem is, you cannot know this programmatically as you must use
+locally defined subtype, which we have to hard-code, i.e. discovery
+needs to know each and every platform statically.
+
+Today you can get LLDP implementation from vendor which is useless for
+automated discovery, as you cannot guarantee to get useful information
+how to connect to peer device nor useful information how to
+discriminate correct interface at peer.
+Wouldn't this be goal #1 for L2 discovery? That automated discovery is possible?
+
+Would it be too complex to mandate that if system implements SNMP and
+IPv4 or IPv6, it must send networkAddress (and define it as address
+which responds to SNMP) and chassisID with
+new ifIndex subtype? This way all real-world IP devices could be
+programmatic discovered without vendor specific code to support it.
+You'd still not require things which might be hard to implement in
+some niche/embedded situations.
+
+Why can't we have multiple subtypes advertised per chassisID and portID?
+
+
+Also I see that ifAlias is often offered, and two major vendors both
+have had bug in LLDP implementation where portID is ifAlias, which
+mean (against MIB) that they are sending port description (which is
+useful, as it typically only talks about the peer, not local names).
+And vendors happily accept it is bug (while according to standard it
+is not) and have fixed it.
+
+                                           MIB           Real world
+Interface description  ifDescr            ifAlias
+Interface long name   ifName           ifDescr
+Interface short name  ifAlias             ifName
+
+Consequently leaving ifDescr out as valid subtype means in real world
+you don't deliver what you intended to deliver.
+(It is must unfortunate real-world has 'wrong' values in them, but
+that historic problem probably won't be fixed ever, easier to fix MIB)
+
+--
+  ++ytti
+
+===
+Unsubscribe link: mailto:[email protected]
+IEEE. Fostering technological innovation and excellence for the benefit of humanity.
+```