Convolution is a great illustration of the difference between the push and pull approaches. You can implement (one-dimensional, but generalizes easily to any di

Convolution is a great illustration of the difference between the push and pull approaches. You can implement (one-dimensional, but generalizes easily to any dimension) convolution either by assigning each point a weighted average centered at that point, computed over all points (pull), or by sliding the weight function along the interval and accumulating (summing) the weighted value of the current center of the weight function at each point (push). The push approach is optimized for reads (linear reads, quadratic writes), and the pull approach is optimized for writes (quadratic reads, linear writes).