Good Wörishofen (part four)

I BUILT a substantial advantage from the opening in my round three game, only to choose a fundamentally flawed middlegame plan.

How should White proceed (in general terms - no need for a specific line) after Black played 20...Rc8-a8 in Spanton (1901) - Alfred Schmidt (1784)?

Visually, White is much better, and indeed analysis engines reckon White's positional advantage is worth about a pawn or more.
I played 21.f4, but this may throw away most, if not all, of White's edge. Komodo9, for example, continues 21...a6 22.Nc3 Bf6, reckoning the position to be equal.
Looking at the diagram position in the light of the analysis engines' verdict, I believe the point is that although White has a beautifully posted knight on f5, there is little chance of a mating attack as Black has so many pieces clustered around his king.
This is why the engines reckon White maintains a substantial advantage by playing on the queenside, for example with 21.Bd2 or 21.a4.
In the game, AS "believed" my kingside attack, meeting 21.f4 with 21...f6?, which horrendously weakens the light squares around Black's king.
After 22.h5, the engines reckon White is winning, and I did indeed eventually crash through on the kingside.