We have covered the three hub-database models, so let's discuss how to decide which model to use. The repository model is the most attractive, because it provides a real source of master data that is always current and consistent. The other choices involve data replication, so there is usually some latency between data updates and hub updates. Master data is generally fairly static, so a little latency is not necessarily unacceptable. The non-repository approaches also maintain multiple copies of some data, so consistency (keeping the copies the same) is an issue these approaches must deal with.
The downside of the repository model is that it can be extremely expensive and take a long time to implement, because it requires changes to the applications that maintain and consume the master data. The repository model makes sense if: the number of applications involved in the MDM project is limited; you have enough control over the applications to make the required modifications; and the availability of authoritative and consistent master data provides enough business value to justify the time and cost required to build a repository-model MDM hub.
A registry-model MDM hub is appropriate when only a limited number of non-performance–critical queries involve access to a significant number of the application databases integrated with the MDM hub. Registry-model hubs are cheaper and quicker to implement and can be implemented one data source at a time, so they are good for incremental implementation and provide early return on investment (ROI). Registries are not good when queries routinely return attributes from many application databases or when there is enough duplication of data, so that determining which of several alternate sources of an attribute to return is a complex decision. In these cases, the pre-integrated and cleansed data provided by a hybrid-model MDM hub provide a more efficient and consistent source of master data.
It's important to note that the hybrid model is not a single model, but instead a whole continuum of options that start at the registry model and continue through to the repository model. For this reason, you may decide to start with a solution close to the registry model and gradually expand the number of attributes integrated into the MDM hub until you have an MDM repository implemented. Because MDM projects can be very expensive and time-consuming in a large enterprise with many applications, it's good to have a strategy that allows you to implement incrementally by both gradually increasing the number of attributes stored in the hub and incrementally adding applications to the hub. This allows you to show an early ROI from the MDM project, with a clear path to a long-term enterprise-wide solution.
No comments:
Post a Comment