A number of weeks in the past, Daniel Dib tweeted a slide from Radia Perlman’s presentation wherein she claimed IPv6 was the worst resolution ever as we might have adopted CLNP in 1992. I had comparable ideas on the subject a couple of years in the past, and over tons of discussions, weblog posts, and creating the How Networks Actually Work webinar slowly realized it wouldn’t have mattered.
Might we use CLNP because the next-generation IP? After all, and we might have saved a couple of years – CLNP was already a mature protocol. Sadly, it turned out even IETF wasn’t pragmatic sufficient to acknowledge superior expertise from what some thought-about their arch-enemy. You’ll discover a couple of juicy particulars in The Parts of Networking Model by Michael A. Padlipsky.
No matter that not-so-brilliant episode in IETF historical past, I inevitably get upset when CLNP aficionados use bogus arguments to justify their gripes. The “Worst Choice Ever” slide accommodates a bunch of them, beginning with these:
CLNP wouldn’t want ARP. There’s completely no method round having a protocol mapping layer-3 addresses into layer-2 addresses except you’re constructing your community with point-to-point hyperlinks that don’t want layer-2 addresses by definition.
CLNP simply used a special strategy than IP. As a substitute of a demand-driven ARP (ask while you want the knowledge), CLNP makes use of ES-IS (periodic hellos). We might spend method an excessive amount of time discussing whether or not one is best than the opposite, and it doesn’t matter within the days of 4-core 2GHz CPUs. Simply don’t declare that CLNP is best than IP as a result of it doesn’t want ARP.
CLNP wouldn’t want Ethernet. The actual motive Ethernet and different shared-media LAN options turned common was decrease value. You want two interfaces per hyperlink in any switched or routed answer. You solely want one while you’re attaching endpoints to a shared medium. The scenario turns into much more ridiculous when the endpoints require excessive bandwidth consumed in bursts, protecting the typical endpoint interface utilization low.
What mattered in Nineteen Eighties clearly nonetheless issues in 2020s: when each information heart switching ASIC consists of line-speed layer-3 forwarding, the service suppliers are nonetheless constructing Passive Optical Networks (PON) as a substitute of switched Ethernet or routed IP networks.
One more reason for the explosion of bridged options was the simplicity of the protocol stack. You needed to implement STP – which might simply match into the 64 KB handle house of an 8-bit CPU – as a substitute of a dozen community layer protocols and a bunch of related routing protocols.
Lengthy story brief: Utilizing CLNP as a substitute of IPv6 wouldn’t change a factor. We might nonetheless need to use lower-cost connectivity options (Ethernet), and distributors making an attempt to implement a networking stack with minimal funding would nonetheless create crappy bridges or spaghetti mess of static routes and NAT.
