[ih] TCP fat pipe chronology (was Ken Olsen's impact on the Internet)

[Craig Partridge]

> Craig et al.,
>    With regard to TCP and fat pipes, there were (at least) two kinds of 
> things going on:
> <> changes to algorithms while leaving the TCP protocol untouched (e.g., 
> improved retransmit timers and VJ's wonderful congestion window work), and

DEC (aka Raj Jain and KK Ramakrishnan) tumbled onto the timer and congestion
problems about the same time as Van, Phil Karn, Lixia Zhang, and I did.

Raj and Lixia both wrote papers in 1986 describing the retransmission ambiguity
problem that Phil and I solved in 1987.

Raj and KK did work showing that additive increase and multiplicative decrease
was the right congestion approach concurrently with and independent of
Van's work.  Raj and KK's paper was published in the same session at
SIGCOMM '88 as Van's work (still one of my favorite SIGCOMM sessions of
all time).  Raj and KK's approach (the DECbit scheme) went into DECnet
Phase-IV.

Van was the only one to figure out that a better round-trip time estimator
was required and how to do it.

> <> eventual changes in the TCP protocol (e.g., window scaling)

To my knowledge, only the Internet community ended up working on the
window scaling and extended sequence number problem.

>    In this context, how did DECnet Phase-IV fit in?  Was it more 
> capable, less so, or about the same as TCP?  I know the HEPnet and SPAN 
> folks were making heavy use shipping (what then passed for) large files 
> around the world.

DECnet Phase-IV was a pretty formidable protocol suite and was deployed at
a scale that one could learn a lot from its protocols.  Remember that
the networking community first discovered the unintentional clock
synchronization problem in large DECNET installations.

Indeed, if memory serves, DECNET was the protocol suite that forced us
to really understand the challenges of the coexistence of multiple protocols
on a single network (e.g. same Ethernet and same long-haul links).

>    Separate question: how would OSI (=?? DECnet Phase-V??) have compared?

My recollection (I have documentation to back it up and thus easily could
be wrong) was that DEC Phase-V was going to be a step backward.  I recall
DEC folks lamenting all the work to retrofit DECNET Phase-IV features
that they'd painfully learned were needed into Phase-V and worrying
about how to get OSI to adopt the improvements.

Thanks!

Craig