Re: Quality of Index

Joe, Since the index is unique and non-composite, each index item will appear as an entity with the rowid next to it. There will be no "grouping" for the common parts of the key. A couple of basic performance tips. 1) on hp, you will get your best performance if you avoid dynamically allocated shared memory segments. Do an onstat -g seg. If you see more that one virtural shared memory segment ("the one with the 'V'"), then your performance will suffer. You need to increase the size of the initial virtural memory segment (SHMVIRTSIZE) so that you are not allocating dynamic segments. 2) run "select * from syssqexplain order by sqx_estcost desc" from the sysmaster database. This will display your current queries in the engine by cost. By studying that, you will be easily able to see where your costs are being incurred. DO NOT RUN THIS ON ANY ENGINE PRIOR TO 7.13!!!!!!!!! Subject: Quality of Index From: jlumbley@netcom.com (Joe Lumbley) Date: Wed, 5 Feb 1997 23:29:59 GMT Message-ID: <jlumbleyE55LA0.5D7@netcom.com> Folks, Running 7.13 on HPUX 10.10 I'm getting some slow performance. One thing I'm looking at is an index on a big table. I remember reading (maybe even writing) about a quirk about Informix indexes. First, the index is a single (non-composite) index on a char(10) field. The entries are unique, but very similar, like this: 09142630000 09142630001 09142630002, etc. There's a **lot** of duplication in the first six positions of the index. My fading memory seems to remember something about the way that Informix groups the keys together before building the index. Am I just making this up? Extra points if someone refers me to some of my own writing with a RTFM ;) Joe -- -------------------------------------------------------------------------- - Joe Lumbley(jlumbley@netcom.com) -------------------------------------------------------------------------- - Madison Pruet