Notes by Thorsten von Eicken, Jan 1999
The correctness of an operation needs to be checked end-to-end. Correctness checks for intermediate results do not help checking overall correctness, however, they may improve performance by avoiding a complete end-to-end retry.
Nice MIT gateway byte-swapping bug anecdote, "ultimate end-to-end check: compare files with old listings and correct manually"
Problems with the meaning of ACKs.
Duplicate message suppression.
Is TCP's notion of a reliable byte stream useful beyond perhaps improving performance? Examples where TCP's reliability is counter-productive?
Is a FIFO delivery guarantee provided by the network useful?
Authentication argument?