One of the ToC practices is the five focusing steps - which start with finding the constraints, then secondarily ensuring the constraint is working effectively by exploiting it, the third is ensuring it is protected by subordinating the other process steps, the fourth is evaluating the constraint if it is still not providing sufficient throughput and the fifth and final is finding the new constraint of the system.

Note that you don't start by elevating the capacity of the constraint or investing in it till you have ensured it is working effectively and is not being swamped by other processes in the system. The idea is that it should have sufficient capacity but not too much. In fact ToC supports more varability than Kanban but this is another post.

I find this a useful model when looking tuning and performance of a system and find the throttling pattern extremely useful. It is the equivalent to the subordination step in the five focusing steps. It allows us to tune the system performance while acknowledging the constraints of a certain part of the process

From a capacity perspective I think of a system as having a certain 'wattage' which is made up of two factors; the first if the responsiveness of the system components and the other is the ability to handle a number of concurrent requests. A system component that is very responsive and can support a high degree of concurrency offers a lot of capacity (putting scaling aside for now).

Where we find an existing component that lacks 'wattage' we can use the throttling pattern to ensure the system is stable and optimised. This allows time to evaluate the capacity of a system and invest wisely.

What is the throttling pattern - Microsoft's take
Overview of the five focusing steps - Kelvyn's excellent overview