[I accidentally saw this in the source list, and see that I ought to 
join the networking list.]


Jonathan Stone wrote:
> Well, if you like, you could always ask Bob Braden.  But in context
> (the rest of RFC-1122, and the requirement to provide a way to disable
> it), the meaning of SHOULD implement as "on by default' seems
> eminently clear to me.

Well, by my (mail) count, this topic has come up 70 times during 1998 - 
today on the end2end list, and 16 times on the tcp-implementers list. 

Supposing that everyone here understands that Nagle is to avoid "being 
nibbled by mice" (many tiny packets congesting the Internet), and 
delayed Ack is for flow congestion control (Ack clocking) that only serendipitously reduces mice....

The solution as I remember it from Sally Floyd's presentation is to 
leave both on for all connections, but for high bandwidth flows, 
slow start should send 4 initial packets rather than 2.  There was a 
fair amount of simulation.  I don't remember the paper -- it's 
probably floating around somewhere.

Anyway, I've been out of the loop for awhile, there might be something 
newer.  But, at least make sure that slow start has been updated. 
-- 
William Allen Simpson
    Key fingerprint =  17 40 5E 67 15 6F 31 26  DD 0D B9 9B 6A 15 2C 32