Re: Distributed Join

In article <1995Jun7.130349.10170@eisner> armstrong@eisner.decus.org (Borden Armstrong, United Airlines) writes: >From: armstrong@eisner.decus.org (Borden Armstrong, United Airlines) >Subject: Re: Distributed Join >Date: 7 Jun 95 13:03:49 -0400 >In article <james.lawrence.45.0012036C@EPAMAIL.EPA.GOV>, james.lawrence@EPAMAIL.EPA.GOV writes: >> However anyone considering this for production systems should be aware >> that nobody is optimizing distributed joins yet. Unfortunately this >> means that there is no value in making it transparent, since it wouldn't >> perform worth a shit. >Not entirely true. Examine Digital's DBIntegrator. And since DBI can >front-end both Sybase & Oracle7 (as well as DB2, Oracle Rdb and a number of >other databases) it is possible to perform optimised distributed joins across >database platforms, not just across databases. >Disclaimer: I do not work for Digital, but I did beta-test the product. >Borden Armstrong United Airlines - SFOKI >Sr. Staff Analyst - Database Design San Francisco International Airport >Armstrong@Eisner.Decus.Org San Francisco, CA 94128 I remain very sceptical. To optimize a join across multiple vendor databases requires that your 'joining' product must fully familiar with all the individual optimizers in each database to understand how they will process their portion of the query. Next it must be familiar with each vendors 'statistics' and translate those into a generic set for use in it's own optimization. (This also means significant maintenance for the product since evryone's statistics and optimization are proprietary and there are no associated standards) Finally it must have a place to store relative communications costs to each node and then consider that during optimization. Fact is most of the vendors are not doing all this within their own databases yet never mind a multi-vendor join. It's realtively simple to do a distributed join, quite another thing to optimize it! Lawrence....... Lawrence.....