For example, in the Internet's current architecture, the decision to divide functionality into particular layers (Physical/Link/Internet/Transport/Application) and the services those layers provide reflect architectural choices. So does its naming, such as the design of IP's addresses - what they mean (i.e., identifying a network interface) and how they are structured (number of bits; prefix hierarchy to facilitate routing; multicast and broadcast functionality; blocks reserved for private networks; no enforcement of source address validity).
For another example, the architecture of TCP is that it provides the abstraction of (1) byte-stream based communication between a pair of peer processes, (2) that is reliable in the presence of packet loss, reordering, duplication, and (3) stable in the presence of congestion. Its mechanisms include the use of sequence numbers, acknowledgments, a 3-way handshake for reliable connection establishment and option negotiation, and a transmission-rate control algorithm for managing congestion. Peer processes are named by IP address / port pairs (thus, the architecture does not support migration of connections when a system's IP address changes).
Architecture often has significant implications for how well the network can perform various tasks, including security considerations. As you do this assignment, keep this notion in mind. Important: try not to overly focus on specific mechanisms used to achieve architectural functionality. The heart of an architecture is its abstractions.
Read the paper
Ethane: Taking Control of the Enterprise, Martin Casado et al., SIGCOMM 2007
Feel free to think boldly! In particular, you should not worry about ensuring backward compatibility or incremental deployment (though you should identify whether or not your scheme has these properties in part (7) below).