Descriptions of implementation size

We can roughly divide implementations into small, medium, and large.

A full definition of these terms includes an accounting of record size (number and size of properties and dimension values per record), total data set size, the number of indexing and MDEX Engine servers, and other measurements of scale.

Although that level of detail is necessary for sizing a specific implementation, it is not necessary for the more general discussion of hardware architecture here. For simplicity's sake, this chapter uses the terms small, medium, and large as follows:

A small implementation means the Dgraph runs an application's data set on a single processor.
A medium implementation means a single Dgraph is mirrored several times for throughput (rather than solely for redundancy), and it means a dedicated server may be necessary for crawling or indexing.
A large implementation means a data set must be partitioned into multiple Dgraphs (that is, an Agraph implementation) and a dedicated machine is required for crawling or indexing.