[net.unix-wizards] Why datagram service is unreliable

buck%nrl-css@sri-unix.UUCP (01/18/84)

From:  Joe Buck <buck@nrl-css>


The design decision that datagram service is unreliable was made
by DOD, in the TCP/IP protocol. This is what 4.2 implements.
Over (say) a local network, the chances of a datagram being
garbled or lost are very slim. The virtual circuit service (TCP)
is implemented by using the datagram service (IP) with sequence
numbers and acknowledgements; there is, of course, a speed penalty for
this.

Berkeley did not design the IPC software, as previous messages
to this list have explained. They merely integrated it into 4.2.

uucp: ...decvax!linus!nrl-css!buck
arpa: buck@nrl-css

jdd@allegra.UUCP (01/20/84)

    The design decision that datagram service is unreliable was made by
    DOD, in the TCP/IP protocol. This is what 4.2 implements.  Over
    (say) a local network, the chances of a datagram being garbled or
    lost are very slim. The virtual circuit service (TCP) is implemented
    by using the datagram service (IP) with sequence numbers and
    acknowledgements; there is, of course, a speed penalty for this.

Another off-the-wall reply.  The 4.2BSD \Internet/ domain implements TCP/IP;
I'm talking about the \UNIX/ domain.

Cheers,
John ("Can Anybody Here Read?") DeTreville
Bell Labs, Murray Hill

guy@rlgvax.UUCP (Guy Harris) (01/20/84)

Oh, by the way, Berkeley did more than just drop BBN's TCP/IP into 4.2; they
tuned it rather a lot.

	Guy Harris
	{seismo,ihnp4,allegra}!rlgvax!guy