Beginning down the path of starting a project like this is complex and unfortunately isn't helped by some of Drupal's (confusingly named) features which describe aspects of reusability but aren't in themselves a full fledged approach to architecting such a network. In addition to that, there are many misconceptions about Drupal's capabilities and affordances when it comes to building such networks.

In order to try and expose some of the discovery and decision making process behind starting an ambitious Drupal network project, the following is a non-exhaustive list of popular architectural paradigms that exist, evaluated on the following axis:

• Up-front investment: the up-front cost of starting a network of sites.
• Per-unit investment: the cost of introducing a new site to the network
• Flexibility: the ability to customise and create bespoke experiences within each network site
• Platform maintainability: the ability to iterate and evolve the network as a whole

As with all complex projects, there are a large number of requirements and constraints which factor into technical decision making, so these approaches are a broad view of the landscape of Drupal's capabilities.

Models of networked sites

Starter-kits

A starter-kit consists of creating a Drupal distribution or install profile, catering to as much common functionality as possible across the network in an initial development phase and then allowing each new website to make any additional required customisations as needed. These customisations may consist of writing additional code, enabling new dependencies and modifying the configuration shipped with the initial distribution.

For each individual site, this model affords the most flexibility. By allowing each site to evolve independently any new requirements or features perceived as bespoke can be implemented and deployed without making consideration to the starter-kit itself or other websites within the network.

The major drawback of this approach is being able to maintain and evolve the network of sites as a whole. Each new site in the network creates a new deployment with it's own configuration, dependencies and code, meaning new features and bug fixes can't be deployed across the whole network without specific individual effort and conflict resolution for each site. In practice once a site is live under this model, it can effectively be considered a siloed project without a significant relationship to other sites in the network.

As far as how feature rich an initial starter kit is largely depends on the project. For example, early versions aGov 8, the starter-kit distribution PreviousNext built and maintained for Australian government organisations was intentionally fairly rudimentary in the amount of content types it shipped with. The goal was a foundation to launch you into best practices, without being overly prescriptive. When future iterations of aGov were released that baked in Drupal 8's new media capabilities, it was not possible to deploy this iteration to all existing installations.

In a similar vein, I would classify govCMS8, the Australian governments own Drupal distribution under this same model. By default, the distribution ships with a lot more features and a lot more ready to go configuration than most starter-kits, however both SaaS and PaaS deployments of govCMS allow a wide scope of deep structural configuration changes to Drupal, which essentially sets up each project to function as a standalone unit after the initial build.

Products

Another less widespread approach is the product model. Under this model, a Drupal distribution is leveraged as the full feature set for all sites in the network and all sites make use of the initial and ongoing development roadmap of the product. This approach is arguably less flexible, since each individual site doesn't have unfettered access to extend and manipulate the platform.

The major advantage of this approach is, a single team can scale their ongoing enhancement and maintenance efforts to the entire network of sites regardless of the number of sites in the network.

Under the product model, since the feature set running on each site is a known and strictly defined set of configuration and dependencies, a team could feasibly migrate hundreds of sites to using new features of Drupal by working on a single software product. All sites would be the target of all new evolutions of the platform and benefit from it's ongoing maintenance. Evolutions of the platform would not strictly be limited to features, but also updating dependencies or moving to new major versions of Drupal. 

An example of a project PreviousNext delivered under this model was a video content platform for group of media organisations. Each organisation could leverage the product to spin up a website to share their videos and engage with their audience. New features were added regularly and at its height, 26 different websites serving vastly different audiences of users would evolve together. One of the challenges of maintaining a product distribution is the governance around flexibility and change. While tempting to allow each site to develop its own product roadmap, when each site required it's own bespoke features, the process for such requests would be followed:

• The site owner raises a request a change to their website, "please replace the hero image on the homepage with a slideshow of images".
• The team evaluates the request and places it on the product roadmap.
• Instead of replacing all hero images with slideshows, the feature is developed an optional choice for content editors: you may either upload slides or a hero image.
• The feature would be built, tested and deployed to all sites in the network.
• The site owner is then able to upload slides and all other site owners in the network have the same capability.

This approach certainly takes a level of control and focused organisation effort to accomplish, however leveraged properly can have significant payoffs for the maintainability of a network of sites as a whole. Examples of product based distributions in the Open Source Drupal ecosystem are Open Social or Drupal Commons.

Federated back-ends

Another approach to building out a network of sites is the notion of a federated back-end. This dovetails with terms like "service oriented architecture", "multi-tenancy" or "content hub", where a single deployment and instance of Drupal is leveraged as a large repository of content for multiple web front-ends.

Under this model, instead of the boundary between users and content being defined by different databases and deployments of Drupal, they must instead be implemented in the application layer. That is, Drupal itself is customised and tailored to meet the access needs of the organisations sharing the same Drupal site. While this is certainly additional work and complexity, if a single group of content authors is responsible for content across the whole network, it can be advantageous to lower these barriers. Maintenance for the content hub is also fairly light touch under this model, since only one installation needs to be updated and maintained.

This pattern also intersects with the "product" style of application. Since all web properties are powered by the same underlying application, they closely share a feature set and product roadmap. While this model is also often deployed in conjunction with a decoupled front-end, Drupal sites are capable of delivering integrated front-ends to network sites from a single federated back-end. In some cases, the federated model has an elevated risk of being a single point of failure, given a single deployment and instance is responsible for sites in the network.

An example of the federated back-end model can be illustrated in the "Tide" component of vic.gov.au's "Single Digital Presence" project. Drupal 8 is deployed as a single instance serving multiple decoupled front-ends. The features of the single content repository are documented and available for evaluation by prospective users. 

Independent sites

One option that isn't often considered in of a lot of organisations when evaluating reusability of features and effort across a network of Drupal sites is simply building multiple completely unrelated sites and using smaller units of functionality as a mechanism for reuse. Drupal has a mature concept for sharing functionality across Drupal sites: the humble module.

While this approach in general doesn't strictly fit the theme of this blog post, in some cases writing and publishing a general case module which doesn't boil the ocean on dictating the tools, approach and features used to build a new Drupal site, is the best solution for some projects.

These kinds of projects would be driven by websites that are characterized as mostly unique with various related and unrelated feature sets. This is also an approach consistent with day to day of Drupal development outside the scope of network projects. With Drupal's open ecosystem, opportunities for collaboration and reuse often drive building functionality in reusable modules and publishing those modules on drupal.org.

Common misconceptions

Install profiles

Drupal "profiles" or "distributions" are not a silver bullet for developing and organising an architecture for networks of Drupal sites. They are incredibly flexible, so how they are deployed and leveraged still contain the same governance and architectural decisions discussed.

Configuration management

Configuration management is not a silver bullet. While configuration management has simplified a range of complex deployment problems that were present in Drupal 7 sites, it doesn't drive a particular architecture for networks of sites. The tools in Drupal core are continually getting sharper and while innovations in contrib have experimented with new mechanisms for deploying and updating configuration, it's not an end-to-end architectural solution for network sites.

Multisite

The multisite concept in Drupal is frequently misunderstood as an approach for building a network of sites. In reality, multisites are a tool for deploying any configuration or setup of Drupal sites to a shared document root. It doesn't produce any tangible outcome as far as project architecture is concerned beyond forcing multiple sites to be hosted on the same server.

Headless Drupal & JavaScript frameworks

While some of these approaches, like the "federated back-end" are significantly benefited by a decoupled front-end, headless Drupal is compatible with all models of network sites. You could build a product or starter-kit that was either fully integrated, progressively decoupled or completely decoupled and the same back-end architectural decisions would apply.

Drupal has strong API based functionality, which can be enabled and configured as required. The approach of evaluating and selecting frameworks or front-ends to consume Drupal, have their own set of considerations that need to be carefully evaluated for fitness in any given project.

Styling and appearance

Styleguide-driven development has largely matured to solve issues with reusability, inheritance and extensibility of visual components. This approach has been the foundation for all new sites built by PreviousNext for last few years, see our blog posts. By building components and style guides, duplication of effort when building front-ends has been minimised across both traditional and network based projects. For that reason the visual style and consistency of sites within a network is not necessarily a factor when considering Drupal architectural paradigms.

Summing up

Given the size of Drupal's ecosystem and the increasingly rapid pace of evolution, describing all of the factors and challenges that play into a large network site project is difficult. As always the process of rigorous discovery and deeply understanding a project's goals and requirements should always be the first step in beginning a technical project.