A graph is Helly if every family of pairwise intersecting balls has a nonempty common intersection. Motivated by previous work on dually chordal graphs and graphs of bounded distance VC-dimension we prove several new results on the complexity of computing the diameter and the radius on Helly graphs and related graph classes. * First, we present algorithms which given an $n$-vertex $m$-edge Helly graph $G$ as input, compute w.h.p. its radius and its diameter in time $\tilde{\cal O}(m\sqrt{n})$. Our algorithms are based on the Helly property and on several implications of the unimodality of the eccentricity function in Helly graphs: every vertex of locally minimum eccentricity is a central vertex. * Then, we focus on $C_4$-free Helly graphs, which include, amongst other subclasses, bridged Helly graphs and so, chordal Helly graphs and hereditary Helly graphs. For the $C_4$-free Helly graphs, we present linear-time algorithms for computing the eccentricity of all vertices. Doing so, we generalize previous results on strongly chordal graphs to a much larger subclass. * Finally, we derive from our findings on chordal Helly graphs a more general one-to-many reduction from diameter computation on chordal graphs to either diameter computation on split graphs or the {\sc Disjoint Set} problem. Therefore, split graphs are in some sense the {\em only} hard instances for diameter computation on chordal graphs. As a byproduct of our reduction the eccentricity of all vertices in a chordal graph can be approximated in ${\cal O}(m\log{n})$ time with an additive one-sided error of at most one, and on any subclass of chordal graphs with constant VC-dimension the diameter can be computed in truly subquadratic time. These above results are a new step toward better understanding the role of abstract geometric properties in the fast computation of metric graph invariants.