The application stays connected to the DB system even when it is not using DB services. This commonly wastes valuable and expensive database resources, as most of the time applications only query and view the persistent data. ADO.Net solves this problem by managing a local buffer of persistent data called data set. 
Our application automatically connects to the database server when it needs to run a query and then disconnects immediately after getting the result back and storing it in the dataset. This design of ADO.Net is called disconnected data architecture and is very much similar to the connectionless services of HTTP on the internet. ADO.Net also provides connection oriented traditional data access services.
An important aspect of disconnected architecture is that it maintains a local repository of data in the dataset object. The dataset object stores the tables, their relationship and their different constraints. The user can perform operations like update, insert and delete on this dataset locally, and the changes made to the dataset are applied to the actual database as a batch when needed. This greatly reduces network traffic and results in better performance.