Creates a default cluster using one less than the total cores available on the system. Where the number of cores cannot be determined a cluster of one is created. By default this uses forking, which is not be available on Windows. Hence, the fork parameter has no effect on Windows.
Details
The workers are given parallel random number streams, so that a windowed
analysis that uses randomness can be run again and give the same numbers.
Without them each worker seeds itself from its own entropy and set.seed()
in the session cannot reach it, so the analysis gives different numbers on
every run. Where seed is NULL the streams are drawn from the session's own
random number state, so set.seed() before the call fixes them, and that
state and the session's generator are left as they were found.
The streams are one per worker, so which worker computes a window decides
which numbers it draws: reproducing a result needs a cluster of the same
size as well as the same seed, and the same result cannot be expected of
windowing() without a cluster, which draws from the session's own stream.
Examples
if (FALSE) { # \dontrun{
cl <- defaultCluster()
stopCluster(cl)
cl <- defaultCluster(FALSE)
stopCluster(cl)
# The same numbers on every run, from the session's seed or from the
# cluster's own.
set.seed(42)
cl <- defaultCluster()
stopCluster(cl)
cl <- defaultCluster(seed=42)
stopCluster(cl)
} # }
