Maintaining scalability and efficiency in conventional networks often requires
huge amount of administrative effort to configure network devices
and protocols. Large configuration overhead increases operational expenditures

and leads to frequent failures due to configuration mistakes.
On the other hand, self-configuring networks, such as Ethernet, work only
on a small scale, because increasing the size of the network rapidly degrades
its performance (e.g., skews link utilization, stretches traffic-forwarding paths,
lowers availability, delays failover, etc.).

This thesis proposes a simple and generic set of architectural primitives
that enable self-configuring in a large network without sacrificing performance.
Those primitives are flat addressing, traffic indirection, and caching of routing state.
By building upon these primitives, the thesis proposes and evaluates three specific
architectural solutions designed respectively for enterprises, data centers, and providers:
i) a scalable Ethernet architecture for large enterprises (SEATTLE),
ii) a next-generation data-center network architecture (Monsoon), and
iii) a scalable VPN routing architecture for large tier-1 providers (Relaying).
The resulting architectures collectively demonstrate the efficacy of the
primitives in resolving the tussle among scalability, efficiency, and lower
management complexity.