Network architecture and protocols

The key to understand complex networks is understanding their architecture. Architecture is the most universal, high-level, and persistent elemments of structure and organization (or princciples of structuring and organizing a complex system). Protocols define how diverse modules interact, and architecture defines how sets of protocols are organized. Architecture usually involves specification of protocols (rules of interaction) more than modules (which obey protocols). In engineering, system architecture must facilitate system level functionality as well as robustness and evolvability to uncertainty and change in components, function, and environment.

Take, Internet, as an example. The Internet is an obvious example of how a protocol-based architecture facilitates evolution and robustness. The architecture of TCP/IP (Transmission Control and Internet Protocols), or the hourglass protocol stack as it’s known, has a thin, hidden “waist” of universally shared feedback control (TCP/IP) between the visible upper (application software) and lower (hardware) layers. The term “hourglass” has been used because there is a vast diversity of applications and hardware that sit above and below the thin waist of universally shared control mechanisms (TCP/IP). Roughly, IP controls the routes for packet flows and thus, available bandwidth. Applications split files into packets, and the TCP controls their rates and guarantees delivery. This allows “plug-and-play” between modules that obey shared protocols; any set of applications that “talks” TCP can run transparently and robustly on any set of hardware that “talks” IP.

Similarly, in the microbial biosphere, genes that “talk” central transcription and translation protocols can move by horizontal gene transfer (HGT), also accelerating evolution in a kind of “bacterial internet.” But also like the technological Internet, the functionality of the resulting new proteins that result is enhanced by when they have additional shared protocols, such as group transfers and carriers in metabolism, and conserved residue pairs in signal transduction. Thus selection acting at the protocol level could evolve and preserve shared architecture, essentially “evolving evolvability.”

In most networked systems, protocols within a layer can be visualized as “bowties,” which have large fan-ins and -outs of energy, materials, and information via a thin ‘knot’ of universal protocols specifying carriers, building blocks, or interfaces. Other engineered examples of bowties include networks which connect energy sources to diverse consumers via carriers and standard socket-plug interfaces, and raw materials to assemblies via standardized building blocks in advance manufacturing. In these and the biologic examples, the currencies, carriers, intermediates, precursors, plugs, packets, conserved residues, and interfaces in the bowtie “knot” are highly constrained by protocols. Yet their shared universality allows diverse and robust edges toadapt and evolve, as long as they have appropriate (and typically hidden) layers of feedback control such as TCP/IP.

While there does not yet exist a coherent theory of architecture, a condition we aim to change, there are now major developement and progress in both technology and biology towards a unified and coherent theory of architecture. Especially, a rich optimization theory has recently been developed to reverse and forward engineer the architecture of the Internet and related technologies. This framework views the network as solving an appropriately defined optimization problem, ranging from the classical network flow problems often formulated as linear programs, to the recent and more general Network Utility Maximization (NUM) problem. Then, network layering can be understood as a decomposition of this large optimization problem into ‘decentralized’ subproblems, and various protocol layers are regarded as carrying out asynchronous, distributed computation to implicitly solve this global optimization problem. Different layers iterate on different subsets of the decision variables using local information to achieve individual optimality. Taken together, these local algorithms attempt to achieve a global objective. Such a theory facilitates both understanding and design of network architectures, and as such could potentially be relevant to problems in both systems and synthetic biology. In reserve engineering a given network, identifying an underlying optimization problem being solved will expose the interconnection between protocol layers and can be used to study rigorously performance tradeoffs in protocol layering as different ways to distribute a centralized computation. In the context of design or ‘forward engineering’, this framework formalizes the common practice of breaking down the desired system into simpler modules, and allows us to systematically carry out the layering process and explicitly trade off design objectives.

We aim to combine and extend these ideas to provide a coherent theory of the layering, protocols, bowties, and hourglasses that make up the architecture of biology.