Given a data set with two numerical features with the aim of computing the Euclidean distance between objects is a time consuming problem. Is there any suggestions about how the key should look like?
You mention a dataset with two numerical features, hence you have points M(x, y) in a plane.
There is a lot of pre-existing knowledge on data sets in a plane: this is called an image!
First you may not need to compute distances, all of them. Use some topology in the plane, with a "tiling".
In other terms you can define quantizers where all points M in set Qi are represented by a selected typical member qi of Qi, assuming that Q0... Qn are non intersecting sets covering the plan P.
The distance between point M in Qi and point M' in Qj can be approximated by dist(qi, qj).
This is just an illustration of many possibilities existing.