A lot of things about SNMP seemed cumbersome to me even 15 years ago. One example is the concept of MIB being a local resource to "make sense" of the otherwise numeric OIDs.
Has SNMP been modernized or mutated into something else? Is it still a must-have feature for network equipment?
Sadly, SNMP is still in common usage. Later versions of the protocol have addressed numerous issues in SNMPv1, but those have almost entirely been directed at fixing the security model. As a result, SNMP traffic is now comparitively bloaty, but they have not addressed what I consider to be the glaring shortcoming in SNMP - that data stored in the MIB resides outside the monitoring/monitored device exchange.
The separation of the MIB-stored data from that exchange, and the consequent use of numeric OIDs on the wire, made sense in SNMPv1, as it kept most exchanges to a single UDP datagram in each direction. As of v3, it no longer makes any sense, to my mind - but I'm not the IETF.
Sadly, SNMP is still a sort of lowest-common-denominator management protocol, and I'm constantly surprised how many devices I see out there where the easiest way to extract monitoring data from them is good old RO-community-string-in-UDP-based SNMPv1.
Edit (2018): because it's so germane, I quote from Geoff Huston's excellent article in the August 2018 edition of the Internet Protocol Journal:
I work on network monitoring systems, and SNMP is still heavily used and relied on for the for the following reasons:
All agree with @madhatter, regarding SNMPv2 not going away any time soon.
But regarding the future, several telecoms NEPs are starting to supply NETCONF interfaces (which are XML based) side-by-side with SNMP and it looks to be gaining momentum at expense of SNMP not just as a replacement for CLIs/TL1 but for FCAPS functionality too that was traditionally done via SNMP.
Also see this old (2013) infoworld article.