Even when you are working with large numbers, most numbers are usually small. Most of the code is probably not dealing with the large things, and a large thing

Even when you are working with large numbers, most numbers are usually small. Most of the code is probably not dealing with the large things, and a large thing may consist of a large number of instances that are individually small. I've personally found it useful to always have concrete numbers in mind. An algorithm or data structure designed for N will probably be fine for N/10 and 10N, but it will often be inefficient for N/1000 or 1000N.