[e2e] TCP un-friendly congestion control

[Sally Floyd]

>>However, if you add occasional packet drops (e.g., from slow-start
>>bursts from the occasional moderate-sized competing flow), the
>>Standard TCP is likely to completely fail to use the available
>>bandwidth in a high-delay high-bandwidth environment.
>
>Yes.  But if the network is a special, controlled network, as often is
>in these cases, congestion loss may be rare.

Yep.  

Though if it is a special, controlled network, one could also just
use a CBR flow, presumeably, and be done with it (with reliability
added on top).  That is, if one isn't competing with uncontrolled
best-effort traffic, then TCP or its variants might not be necessary
at all...

- Sally
http://www.icir.org/floyd/