[ih] Failed Expectations: A Deep Dive Into the Internet’s 40 Years of Evolution (Geoff Huston)

[touch at strayalpha.com]

> On May 31, 2023, at 12:30 PM, Karl Auerbach via Internet-history <internet-history at elists.isoc.org> wrote:
> 
> One of my own observations is how we institutionalize our techno-dislikes.
> 
> ISO/OSI was indeed a disaster, largely self inflicted.
> 
> But it did contain a lot of good ideas.  And we've tended to put distance between the Internet of today and those good ideas.
> 
> My own hobby-horses about things we could have pulled from the OSI carcass are these:
> 
>   - A much better checksum algorithm.  (The OSI Fletcher checksum looks scary but there are ways to compute, anew and incrementally, it that are fast and use only standard 2s complement arithmetic. And it can catch byte reversals that are common when someone forgets to put a hton*() in their code.)
> 
>  - An association (session) layer.  This would have made mobility much easier by allowing an "association" to span multiple transport connections as devices move and change their IP addresses.  It also could have reduced the web's addiction to cookies. (OSI made the mistake of making their session layer excessively complicated,  incomprehensibly documented, and never explaining what it was good for.)

These are just functions that could arguably exist - or be improved - at any “layer”. 

IMO, OSI got layering wrong not just by what iit limited to specific layers, but with the whole idea that layers are absolute in the first place. There’s so much we can do without that constraint, including not only spanning multiple transports, but bridging between packet and circuit models too.

OSI is just the degenerate case of instantiating relative, recursive layers as a particular - and notably arbitrary - fixed instance.*

*(that’s OSI was relegated to one of the last lectures of my intro to networking course)

Joe

—
Dr. Joe Touch, temporal epistemologist
www.strayalpha.com