I still think (without evidence) that rate-based credits are worth
looking into just by the nature of having less control overhead and
being more composable along a path. were there other reasons aside
from an absolute no-drop policy that the digital switches moved from
rate to buffer based credits?
The history was reversed. Digital AN2's per-flow credit scheme
was designed in late 90 before ATM had any congestion control or
even circuit setup standard (these appeared in 93 or so); the original
AN2 had 128 byte cells, a link speed that wasn't a 4^k multiple of 56Kbps,
etc.
Two questions would be:
why didn't ATM standardize on credits given AN2 already demonstrated that
they could be implemented and that they worked? The story I heard was that
ATM believed, on the basis of only simulation and no operational evidence,
that a rate based scheme would be better than a purely credit based scheme
-- credits would have to be augmented with rates, so why
not just do rates? They also didn't want to give Digital a leg up, just
as the industry standardized on 10Mb/s Ethernet in the early 80s to keep
from giving Xerox a leg up (they had already built a 3Mb/s Ethernet).
Part of it was also that they just didn't like Kung, who was pushing credits
at the time (some pretty silly reasons, but remember these are the guys
who brought us 53 byte cells).
A different question was, why didn't AN2 change to rates, since they
changed to cell size and link speed to be compatible with ATM?
In fact, the story they gave was that they would do whatever the ATM
standard was in talking to other vendor switches, but use credits
between DEC switches (in the same fashion as Ipsilon's IP switching).
I think they also felt that ATM's rate based scheme wouldn't really work,
and so ATM would eventually in the long run add credits to the standard.
A similar story applied to circuit setup; DEC had a working system
long before ATM standardized on something else. And the lack of
dynamic circuit setup was a huge hurdle to developing the ATM market.
tom