It relies on the decomposition of the application into smaller services. It goes well with RESTFul microservice architecture or SOA. It is basically a sharing of application code:

The application is sharded into various small services, and each service has its own bound set of responsibilities to perform. A very common approach to decomposing the application is using verb-based decompositions such as login and checkout, and another approach is based on noun-based decompositions such as customer, vendor, buyer, and so on.

We can also mix the approach of the x and y axes, as explained in the following diagrams:

Here, services are broken into sub-services first, and then, each service is scaled on the x axis basis. The second load balancer runs the clone code for each service. The combination of both y and x axes scaling is increasingly a good choice.