Fillfactor behavior
Posted in 1996
It seems that when an index is built, the fillfactor is only approached when
the index is unique, or "fairly close" to unique. Otherwise, you can (and
frequently do) end up with leaf pages as poorly packed as 40%-60%. oncheck -pT
will show you this. The result is I do a lot more index page reads than
necessary, and waste a lot of buffer space. Fortunately I've got plenty, at
least for now.
According to tech support, I'm the second or third person to report this as a
bug, and each time, the problem is closed with comments like "working as
designed, documentation needs clarification". I think it's pretty weak that
an index can't always hit the desired fillfactor at build time. That's one of
the main points of building them bottom up - so you never miss.
Now, I'm well aware of the desirability of unique indexes, and they become less
useful as they become more duplicative, yet they're still a heck of an
improvement over a sequential scan most of the time.
I'd be interested to hear others opinions on this - If I'm all wet, I'll
willingly take my lumps....
--
Greg Moye
EDS Corp. - Plano TX