I assume you are referring to the ISO model described at
and not the DoD model described farther down on the same page. If this
is incorrect, please make a clarification request so I can correct the
At "level 3" (the network layer), the messages are routed based on the
best path which may vary in time. For example, congestion at some
intermediate point may require messages to be discarded, requiring
retransmission if reliability is required. With the current protocols
(e.g., IP), acknowledgement is not required so the overall traffic on
the network can be minimized if reliability is not required.
At "level 4" (the transport layer), the application has a choice
- "reliable" transmission (TCP - all data received in order, error otherwise) OR
- "unreliable" transmission (UDP - a "best effort", no errors)
UDP for example can be used when an application is sending a
"multicast" where the number of receiving systems is unknown prior to
sending the data. To do the same with TCP would require a dedicated
connection between the sender / receiver and much more bandwidth.
So, in short, moving the reliability from layer 4 to layer 3 would
require a great increase in network traffic from at least two factors:
- acknowlegements for messages where reliability is not required
- sequential (not multicast or broadcast) messages between systems
If you are interested, I can look up some data related to the ratio of
TCP vs UDP type transmissions on the Internet.
There are also protocols (e.g., BOOTP) which do not rely on reliable
transmission of data that may not work properly if reliable IP is
required at level 3. In the BOOTP case, it might require configuration
of each booting system to directly address the host server(s).
network protocol levels
reliable unreliable network protocols
percentage TCP UPD