[ih] Installed base momentum (was Re: Design choices in SMTP)

[John Day]

To some degree, Stratacom had the right idea and then ATM screwed it up.

As I explained Yaakov at a SC6 meeting in the mid-90s in Seoul, the right next step was what became MPLS. But they screwed that up too.

By then it was apparent where the problem was going, but there was still a lot of detail to work out. (It takes longer when it isn’t your job.)

> On Feb 14, 2023, at 16:12, Brian E Carpenter <brian.e.carpenter at gmail.com> wrote:
> 
> On 15-Feb-23 02:03, John Day via Internet-history wrote:
>> Yep, thanks.
>> I use to have a slide that had (drawn to scale), 576 byte IP packets over 53 byte ATM cells, over 1500 byte SONET frames with caption, “Why?"
> 
> There was a report by Ramon Caceras** in 1991 which (from memory) showed that based on the observed distribution of IPv4 packet sizes on the Internet, the worst possible cell size, i.e. the cell size that would lead to the maximum number of runt cells with partial payloads, was ~50. In other words, given that the alternative cell payload sizes considered by CCITT were 32 and 64, they chose the worst possible compromise, from the viewpoint of Internet traffic.
> 
> I once sat at dinner with Jacques Saint-Blancat from IBM La Gaude (in France) who was secretary of that particular CCITT study group, and proudly told me that he personally was the first person to write down that "48" payload size in his meeting notes.
> 
> ** http://www.icsi.berkeley.edu/pubs/techreports/tr-91-043.pdf
> 
>   Brian
> 
>>> On Feb 13, 2023, at 16:40, Olivier MJ Crépin-Leblond via Internet-history <internet-history at elists.isoc.org> wrote:
>>> 
>>> 
>>> 
>>> On 10/02/2023 17:56, John Day via Internet-history wrote:
>>>> ATM was a bad idea the day it was proposed.  It was unbelievable how many people were taken in by it.
>>> 
>>> ATM saw glory in PPPoA in ADSL - https://en.wikipedia.org/wiki/Point-to-Point_Protocol_over_ATM
>>> 
>>> As for why TCP/IP won over OSI, my take is that at the time OSI stacks really worked for large mainframe computers. The advent of PCs & desktop machines that could run early TCP/IP stacks killed the OSI stack. Credit goes in particular to:
>>> - KA9Q - https://en.wikipedia.org/wiki/KA9Q (incidentally why is Phil Karn not in the Internet Hall of Fame?)
>>> - Microsoft implementing TCP/IP in Win95
>>> - Novell NetWare TCP/IP support in the early 90s & its NE2000 card (& compatibles)
>>> 
>>> Mainframes lost the game and along did OSI.
>>> Best,
>>> 
>>> Olivier
>>> 
>>> -- 
>>> Internet-history mailing list
>>> Internet-history at elists.isoc.org
>>> https://elists.isoc.org/mailman/listinfo/internet-history