pcp

There's a problem when pmlogsummary encounters float/double
values which have the value NaN, when these values occur only
once or twice in the archive. As soon as these values are
included into the averages, all subsequent arithmetic on the
average returns NaN as well (of course). Ultimately, this
means we do not get meaningful data for an entire archive.
Since there's no situation that incorporating a not-a-number
value into an average makes sense, we can safely ignore the
records which contain such, and still produce a valid result
from pmlogsummary.