A nearly optimal deterministic parallel Voronoi diagram algorithm
This paper improves upon the result described in the 1990 ICALP paper.
It describes an n-processor, O(loglog(n)) time
parallel algorithm which constructs the Voronoi deagram of a set of
n points in the plane. The model of computation is a CRCW
(concurrent-read, concurrent write) PRAM (parallel random-access machine),
capable of exact integer arichmetoc in constant time.
This paper amplifies and improves upon the 1990 conference paper
`Merging free trees in parallel...'.
As previously, the approach uses divide-and-conquer, exploiting
so-called `beachlines' (called `proximity envelopes' in the
conference paper).
In
the conference paper the beachlines were precomputed using
nlog(n) processors (CREW --- exclusive write). The
same effect is produced in this paper by precomputing approximations
to the beachlines using n processors (CRCW).
Thus we use a more powerful model of computation to help
reduce the processor requirement to n.
The paper remarks that our methods cannot be adapted to exclusive-write
machines in the same run-times.