The parallel work is executed on the step 2. But our
parallel method is different from the intuitive idea that
processing all the nodes which is labeled and un-scanned
with all their neighbor nodes synchronous. For example:
Let L = u1, u2, · · · , um is the set of all the
nodes which is labeled and un-scanned. Let P =
(u, v)|u ∈ Landv ∈ N(u), then all the elements in P is
processed synchronous in step 2.
The method above has a drawback that in every iteration
of the algorithm, the set P must be constructed before step
2 and this work can’t be processed by a parallel mod