Marc, That would be totally useless. Think about it this way: You have two decks of cards, one with red backs and one with blue backs, shuffled together. You want to sort the cards by suit, rank and color of back. You put all the cards are face down on the table, so that all you see is red and blue. You can do the sort into two piles by back color easily, but that doesn't help much with the desired goal. You have to turn them over in two piles, then sort each pile by suit and rank, then merge the piles by interleaving them perfectly. On the other hand, if you start with the cards face up, once you have them sorted by suit and rank it is simple to turn one each pair of cards over and observe the color of the back.
A very wise man by the name of Wayne Erickson once told me, in effect, to only use indexes on columns with more than a few distinct values. He held that the overhead imposed by the indexes on the columns with very few values far outweighs any possible benefit. An index-based retrieval is going to work best when none of your values are identical. Emmitt Dove Manager, Converting Applications Development Evergreen Packaging, Inc. [email protected] (203) 214-5683 m (203) 643-8022 o (203) 643-8086 f [email protected] -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of MDRD Sent: Wednesday, January 06, 2010 9:47 AM To: RBASE-L Mailing List Subject: [RBASE-L] - Re: Update command taking forever Is it OK to index a Y/N column if you have 95% N and 5% Y and you only search for Y? I was wondering if those 5% Y's would pop up faster compared to RBase going row by row. Thanks Marc . Do not index any column with relatively few > distinct values (such as a Y/N column). > > Emmitt Dove > Manager, Converting Applications Development > Evergreen Packaging, Inc. > [email protected] > (203) 214-5683 m > (203) 643-8022 o > (203) 643-8086 f > [email protected] > > -----Original Message----- > From: [email protected] [mailto:[email protected]] On Behalf Of Dan > Sent: Wednesday, January 06, 2010 8:19 AM > To: RBASE-L Mailing List > Subject: [RBASE-L] - Re: Update command taking forever > > > Bob, > Thanks for responding. > 1. using latest version dated last week > 2. QUALCOLS is set to 10 already > > Have not unloaded and reloaded in the last week, but that table never > gets deletes, but that still is something to try, can't do it now > till everyone goes home tonight and I stay late. I could try it on > a backup database and see. I guess that is my next step. > > 3. Using the update on just one column where there is a non zero, I > am getting one row updated per second and a half. So the command is > working, but will take a long time. > > Dan > > At 08:08 AM 1/6/2010, you wrote: > >>Dan, >> >>Indexes are only meant to increase the speed of identifying rows, >>not the actual update. >> >>So you definitely do not want to add indexes to the columns you are >>updating. That would >> >>actually slow down the process as you would now be updating the >>indexes as well as the >> >>data. >> >> >> >>Updating 100,000 rows should not take too long, so something is >>certainly askew. >> >> >> >>So some things to check... >> >> >> >>Look at your setting for QualCols >> >> >> >>R:>sho qualcols >> >>QUALCOLS is set to 10 >> >> >> >>If it is not set to 10, then set it. This can make a significant > difference. >> >> >> >>Unload and Reload the database if you have not already done so. If >>there have >> >>been much row deleting etc. sometimes the database files need >>packed. Although >> >>I am hard pressed to think this would cause an hours long update. >> >> >> >>Try updating just one column at a time and see if the time is >>different and report back. >> >> >> >>Make sure you are running the latest version. There were changes >>made that directly >> >>effected speed in certain cases. >> >> >> >>-Bob >> >> >>----- Original Message ----- >>From: "Dan" <[email protected]> >>To: "RBASE-L Mailing List" <[email protected]> >>Sent: Wednesday, January 6, 2010 6:25:51 AM GMT -06:00 US/Canada Central >>Subject: [RBASE-L] - Update command taking forever >> >> >> >> >>Hi, >> We have finally made the conversion to Turbo 8, and am having >>troubles with my month end financial processes. I assumed it was >>index problems and made sure there are indexes. >> >>Update ardetail set invcur = 0, paycur = 0 >> >>used to take seconds in 7.5 26 hours and counting in turbo >> >>in 7.5 neither invcur nor paycur had indexes. >>I added them now, and still no increase in speed. So then I >>thought, ok with the new indexes, lets key just off those.. >> >>Update ardetail set invcur = 0 where invcur <> 0 (excluding 95000 >>rows) only 1300 should now be looked at, and this still takes forever. >> >>Where should I look next? >> > >
