On Fri, Feb 13, 2015 at 3:13 PM, Jim Nasby <jim.na...@bluetreble.com> wrote:
> > If we exclude the issue of needing one or two oddball partitions for +/- > infinity, I expect that fixed sized partitions would actually cover 80-90% > of cases. That would not be true in our case. The data is not at all evenly distributed over the partitioning key. We would need something more like: values a, b, and c get their own partitions and everything else goes in partition d.