🚀 go-pugleaf

In article <8809221350.AA14186@jade.berkeley.edu> "John A. Pershing Jr." <PERSHNG%YKTVMH.bitnet@jade.berkeley.edu> writes:
>No, you're not missing anything.  The reliability provided by the SNA DLC
>layer is carefully preserved by all higher layers, so that additional
>CRCs are probably redundant.  There is probably a tacit "assumption" that
>the various nodes are reliable -- that is, that they won't introduce any
>bit errors without detecting the fault (e.g., via a machine check).

>As I remember (it's been a long time), TCP doesn't make many assumptions
>about the reliability of the lower layers; therefore, it needs some sort
>of checksum to provide reliable transport.  If, in fact, the lower layers
>*are* reliable then TCP probably doesn't need the checksum; however, a
>proper TCP implementation cannot make such an assumption.


Yes, the basic difference is that any transport protocol built on a 
reliable connection (or virtual circuit) can make a number of assumptions
that simplify error control (as is the case for SNA).

TCP is built on top of IP, which is
a datagram protocol (sort of layer 3.5 to the OSI world).  Packets
can arrive out of sequence, in duplicate, in error, or not at all.
Therefore it is up to TCP to provide the reliable end-to-end connection.

An alternate transport protocol in the DoD Internet suite is UDP
(User Datagram Protocol), which is just an interface to the raw IP
level, and does not provide a reliable transport level.

-- 
James Sterbenz  Computer and Communications Research Center
                Washington University in St. Louis 314-726-4203
INTERNET:       jps@wucs1.wustl.edu
UUCP:           wucs1!jps@uunet.uu.net