Reading academic texts on networks, it is easy to get confused by the various labels attached to the networks described in texts. One could say it all began when Paul Baran established the distinction between centralized, decentralized and distributed networks in 1964, in his paper “On Distributed Networks“. This straightforward distinction has been widely adopted by scholars, but there are a few other labels that are often used and yet have a less distinct definition.
Paul Baran (1926-2011) was a Polish-American engineer who invented several technologies related to network architecture. His work on centralized vs decentralized architecture, during the cold war, was inspired by the need to design a network that could withstand a nuclear attack. (More)
One of such labels is that of the federated network, a term often mentioned when talking about the architecture of (alternative) social networks such as Diaspora or Lorea. What is a federated network, though? The word itself hints at similarities with the political concept of a federation (see further on), but how such a state architecture would be applicable to networks is not immediately clear. In this essay, I will take a closer look at this term and how it fits in the network architecture discourse.
As mentioned before, Paul Baran in his classic 1964 text On distributed communication networks distinguishes three kinds of architectures that would be applicable to communication networks. As this text is often cited when discussing network architectures, it is a good place to start with an exploration of network architecture terminology. Baran envisions a network as an interconnected set of nodes, agents that can send and receive data. These nodes can then be organized in various ways, and this organization is what distinguishes the types of network architecture from each other.
- Network: A collection of interlinked nodes that exchange information.
- Node: The most basic part of the network; for example, a user or computer.
- Link: The connection between two nodes.
- Server: A node that has connections to a relatively large amount of other nodes
The “traditional” network architecture is the centralized network, where all nodes send their data to one central node (a server), which then sends the data to the intended recipient. Then there’s the distributed network, where there’s no central server and each node is connected to various other nodes; data simply “hops” through whichever nodes allow for the shortest route to the recipient. Finally there’s the decentralized network, which could be characterized as a distributed network of centralized networks. When “zoomed out”, a decentralized network resembles a distributed network, but zooming in on the nodes of this distributed network reveals that the nodes in this network are in fact centralized networks, with the central node of each network connected to several other such central nodes. This way, the network does not rely on one single server, but splits the risk by having several such central nodes each manage a limited amount of “end point” nodes.
In politics, a federation is a union of states (or other entities) that are partially self-governing and independent but have transferred a set of responsibilities and duties to a central government that unites them.
This is comprehensible enough, but what’s a federated network, then? Alternative social networks such as Diaspora and Lorea have been described as adopting a federated structure, but their server architecture is often strongly remeniscent of Baran’s description of a decentralized system. In Diaspora, for example, users become a member of a “pod”, and in principle only connect directly with their own pod; these pods in turn are connected to each other to allow users in different pods to interact with each other. This matches the “distributed network of centralized networks” description given above quite well.
Is federation just a synonym for decentralization, then? It depends on who you ask. For example, Arvind Narayanan et al., who wrote a comprehensive analysis of the strengths and weaknesses of different network architectures in A Critical Look at Decentralized Personal Data Architectures, would disagree. They distinguish both federated and distributed networks – but mention that these are, in turn, two kinds of decentralized networks: here, “decentralized” is an umbrella term, unlike in Baran’s original text where decentralized networks were clearly distinct from distributed networks. It is clear from Narayanan et al.’s text, however, that their conception of a federated network matches Baran’s description of a decentralized network.
A completely different definition is used by Elijah Sparrow, involved with the Crabgrass alternative social network, in his response to Narayanan et al’s paper. Sparrow uses “decentralized” to refer to peer-to-peer networks, which is what both Narayanan et al and Baran would call “distributed”, meanwhile reserving the label “federated” for “client-to-server-to-server-to-client architectures”, which seems to be another way of describing Baran’s decentralized networks.
Lonneke van der Velden, in her 2013 text Meeting the Alternatives, offers yetanother conceptualization of the relation between decentralization and federation. Writing about the alternative social networks Diaspora and Lorea, she notes that “[decentralization] means that data is not stored on the servers owned by one central actor, but on federated servers”. In this instance, decentralization and federation do seem to be more or less two ways of saying the same thing, where “federation” is a mode of server organization that makes the network of these servers decentralized.
The Social Web
The World Wide Web Consortium (W3C)’s Social Web Incubator Group investigates how, through the use of web standards (such as HTML, OpenSocial or FOAF), an interoperable set of online social technologies can be developed and sustainably maintained.
Looking at another prominent adopter of “federated” terminology, the Federated Social Web W3C community group seems to have yet another conception of what it means for a network to be federated. According to the announcement of their Federated Social Web Summit in 2010, federation (in the context of social networks) is “letting people on different social networks follow each other”. The focus is in this case not so much on network architecture, but rather on data exchange: the FSW proposes that different networks adopt one unified data architecture so that a robust, heterogeneous network-of-networks can emerge. So for the FSW initiative, federation can be achieved through standardization of data formats: the precise connection between nodes and server is less relevant.
Having looked at several different ways of using these labels, where does this bring us? Apparently there’s no single agreed-upon definition of what a federated network is: there’s subtle differences between how the cited authors use this and other labels and sometimes the definitions that seem to be used even contradict each other. This is obviously a problem if one would want to discuss “federated networks” in a general sense.
The most sensible approach might be that of Narayanan et al, who use “decentralized” as an umbrella label that includes both federated and distributed networks. Semantically this is a solid categorization, as both types of network are indeed not centralized. As both federated and distributed networks have “not being centralized” as one of their main distinguishing features, it makes sense to include them in one category based on that characteristic.
Looking at just the “federated” label, what most authors seem to agree on is that a federated network is the same “distributed network of centralized networks” that Baran called decentralized. So in that sense, “federated” has replaced “decentralized” in its original meaning, while “decentralized” has been adopted in various different ways, mostly as an umbrella term. This is a reasonable redefinition, considering the fact that a distributed network could technically also be considered to be decentralized, and the close resemblance of the federated structure to the structure of a political federation, where autonomous sub-entities (such as American states) form one single larger entity (the United States).
The most workable middle ground, then, seems to be that a federated network is a distributed network with each node of this distributed network being a centralized network. Such a federated network is a type of decentralized network, with another type of such a network being the distributed network.
- “FedAll2010/Announcement – Federated Social Web Incubator Group.” Web. 7 Mar. 2013.
- Baran, Paul. “On Distributed Communications Networks.” Communications Systems, IEEE Transactions on 12.1 (1964): 1–9. Print.
- Narayanan, Arvind, Vincent Toubiana, Solon Barocas, Helen Nissenbaum & Dan Boneth. “A Critical Look at Decentralized Personal Data Architectures.” arXiv preprint arXiv:1202.4503 (2012): n. pag. Web. 26 Feb. 2013.
- Sparrow, Elijah. “Appendix to ‘Unlikely Outcomes’ – Unlike Us Reader.” Unlike Us. Web. 7 Mar. 2013.
- Velden, Lonneke van der. “Meeting the Alternatives: Notes About Making Profiles and Joining Hackers.” Unlike Us Reader. Ed. Miriam Rasch & Geert Lovink. Amsterdam: Institute of Network Cultures, 2013. 312–322. Print.