From: Marty Himmelstein Subject: Re: Oracle HINTS and performance Date: 1995/08/08 Message-ID: <406jiv$ceo@dartvax.dartmouth.edu>#1/1 references: <405qjo$mug@data.interserv.net> content-type: text/plain; charset=us-ascii organization: ValleyNet mime-version: 1.0 newsgroups: comp.databases.oracle x-mailer: Mozilla 1.1N (Windows; I; 16bit) deangup@admin.ci.seattle.wa.us wrote: >I tried a hint to use a multicolumn index, ran an EXPLAIN, and discovered >that it had added another layer of nested joined table scans instead. I am >not enthusiastic about hints anyhow, because I am concerned about building >such procedural directives into a nonprocedural language, it seems to subvert >the data independence of the program. > Relational optimizers haven't earned nicknames such as pessimizers or randomizers without just cause. Especially for large decision support applications, it's optimistic to expect the optimizer to always get it right, or even close. Anyway, getting Oracle to run correctly takes time, energy and expertise, and the Oracle tuning techniques are different than those for Sybase, Informix, etc. Where's the independence in that? Oracle, at least, is pragmatic about hints. Sybase has hints, too, (less extensive )but they are officially undocumented. Microsoft documented the hints for their new version of SQL Server, based on the Sybase database. Regards, Marty Himmelstein marty.himmelstein@valley.net