Piethein Strengholt, author of Large-scale data managementrecently published an article presenting six data mesh governance topologies and domain granularity. Each topology adapts the data mesh strategy to balance requirements such as data ownership, organizational structure, rate of change, technology, etc.
Data mesh is an enterprise data architecture that adapts and applies learnings from building distributed architectures to the data domain. Data mesh recommends building a self-service data infrastructure, treating data as a commodity, and organizing teams and architecture according to business domains.
According to Strengholt, organizations moving to data mesh have to make trade-offs. Depending on their context, they might favor quality over control, simple structures over complex structures, etc. Organizations should adapt the data mesh concept to implement what works best for them today and use one of the following topologies as a starting point:
Applying the purest theoretical form of data meshing seems ideal, but it is difficult to achieve because it requires standardization and dealing with issues of capacity duplication, network access, etc. It is often used in cloud-born organizations, which are relatively young and have highly skilled software engineers.
Organizations that value quality and compliance over agility can implement the thin, fully-governed mesh topology because it adds a central distribution layer to solve distribution problems. In Strengholt’s experience, many organizations, especially financial institutions and governments, tend to use it. However, organizations with legacy systems that are difficult to maintain or lack highly skilled software engineers can implement the “hybrid federated mesh” topology where federation occurs on the consumer-side domains and centralization in the others.
Organizations in the supply chain management, product development or transportation sectors require a high level of specialization. Value stream aligned mesh topology solves this scenario by grouping domains that work together. Each value stream will behave as a larger domain, and crossing its boundaries requires data products to adhere to centralized standards.
Large-scale organizations with complex architectures and many applications require multiple levels of governance, alignments and decompositions, Strengholt wrote. Implementing a data mesh will require more trade-offs. In coarse-grained aligned mesh topology, domains contain many applications and boundaries are defined by reflecting business units or regional viewpoints. This topology contradicts a pure data mesh implementation and introduces new risks like the creation of silos.
Finally, some large-scale organizations may implement coarse-grained governed mesh to overcome complexity, peer-to-peer distribution, and interoperability gaps. This topology adds a central distribution layer and relaxes controls across larger boundaries.