Connecting to Another Dataset

There are specialized client datasets that use the BDE or dbExpress to connect to a database server. These specialized client datasets are, in fact, composite components that include another dataset internally to access the data and an internal provider component to package the data from the source dataset and to apply updates back to the database server. These composite components require some additional overhead, but provide certain benefits:

Client datasets provide the most robust way to work with cached updates. By default, other types of datasets post edits directly to the database server. You can reduce network traffic by using a dataset that caches updates locally and applies them all later in a single transaction. For information on the advantages of using client datasets to cache updates, see Using a client dataset to cache updates.
Client datasets can apply edits directly to a database server when the dataset is read-only. When using dbExpress, this is the only way to edit the data in the dataset (it is also the only way to navigate freely in the data when using dbExpress). Even when not using dbExpress, the results of some queries and all stored procedures are read-only. Using a client dataset provides a standard way to make such data editable.
Because client datasets can work directly with dedicated files on disk, using a client dataset can be combined with a file-based model to allow for a flexible "briefcase" application.

In addition to these specialized client datasets, there is a generic client dataset (TClientDataSet), which does not include an internal dataset and dataset provider. Although TClientDataSet has no built-in database access mechanism, you can connect it to another, external, dataset from which it fetches data and to which it sends updates. Although this approach is a bit more complicated, there are times when it is preferable:

Because the source dataset and dataset provider are external, you have more control over how they fetch data and apply updates. For example, the provider component surfaces a number of events that are not available when using a specialized client dataset to access data.
When the source dataset is external, you can link it in a master/detail relationship with another dataset. An external provider automatically converts this arrangement into a single dataset with nested details. When the source dataset is internal, you can't create nested detail sets this way.
Connecting a client dataset to an external dataset is an architecture that easily scales up to multiple tiers. Because the development process can get more involved and expensive as the number of tiers increases, you may want to start developing your application as a single-tiered or two-tiered application. As the amount of data, the number of users, and the number of different applications accessing the data grows, you may later need to scale up to a multi-tiered architecture. If you think you may eventually use a multi-tiered architecture, it can be worthwhile to start by using a client dataset with an external source dataset. This way, when you move the data access and manipulation logic to a middle tier, you protect your development investment because the code can be reused as your application grows.
TClientDataSet can link to any source dataset. This means you can use custom datasets (third-party components) for which there is no corresponding specialized client dataset. Some versions of Delphi even include special provider components that connect a client dataset to an XML document rather than another dataset. (This works the same way as connecting a client dataset to another (source) dataset, except that the XML provider uses an XML document rather than a dataset. For information about these XML providers, see Using an XML document as the source for a provider.)

There are two versions of the architecture that connects a client dataset to an external dataset: