[ih] TTL [was Exterior Gateway Protocol]

[Jack Haverty]

IMHO, it's unfortunate that TTL has officially become just a hop
count.    That was just the best we could do then.

At the time, circa 1982/3, Dave Mills had been showing some pretty
impressive (to me at least) results, being able to achieve
millisecond-level synchronization among his Fuzzballs, using only
inexpensive on-board clocks with some complicated algorithms to interact
with an atomic clock somewhere.   NTP still does that today, and all the
computers in my house know what time it is thanks to that work.   Yours
too probably.

The plan was to incorporate such inexpensive clocks into gateways, use
Dave's NTP magic to achieve an appropriate precision of sync, and make
TTL and routing actually based on time rather than hops.  Among other
things, that would have enabled a true "fastest possible delivery"
datagram service, selected by a setting in the TOS and TTL fields for
traffic flows that needed it.  Seemed pretty straightforward at the
time, but I guess it didn't happen.   I don't see how "hops" are very
useful now; does anything in the net ever use that information for
purposes other than to blow datagrams out of stable orbits?

Today, as I watch all the talking heads and interviewees on TV
pixelating in garish color schemes with hauntingly garbled audio, I
wonder if it has something to do with the TTL/TOS part of Internet
History.   I keep looking behind the TV to see if the bit-bucket has
overflowed.

/Jack

On 9/5/20 3:21 PM, Joseph Touch wrote:
>
>
>> On Sep 5, 2020, at 3:06 PM, vinton cerf via Internet-history
>> <internet-history at elists.isoc.org
>> <mailto:internet-history at elists.isoc.org>> wrote:
>>
>> no disagreement about discard - just that TTL wasn't interpreted as time
>> but hop count in the implementations with which I am familiar;
>
> That’s the spec requirement since RFC 1009 (1987), i.e., decrement by
> as many seconds as the packet is held, but never less than 1. For most
> routers, that effectively means TTL equates to hop count (few queues
> build up by seconds, except in more recent bufferbloat examples -
> because, AFAICT, before bufferbloat the cost of memory made such long
> queues rare relative to link speeds).
>
> IPv6 made that change complete, calling the field “hopcount” anyway.
>
> Joe
>
>
>