Hello all - I'm working on a postgres project after coming from a MySQL background ( no flames, please :). We are importing fairly large xml datasets ( 10-20 MB of xml files per 'project', currently 5 projects) into the database for querying. We are using PHP to create a web interface where users can create projects and upload their files. We have a parsing routine that gets the xml data into the database. The parsing has turned out to be pretty intense. It takes about 10-20 minutes for any project. When we are parsing data, it really slows down the site's response. I tested serving static webpages from apache, endless loops in php , but the choke point seems to be doing any other query on postgres when constructing a php page during parsing. As an example, the original designer specified separate tables for each project. Since they were all the same structure, I suggested combining them into a single table with a project_id column, but he said it would take too long to query. I was suspicious, but I went with his design anyway. It turned out he was right for our current set up. When I needed to empty the project table to re-parse data, doing a cascading delete could take up to 10 minutes! I cut re-parsing time in half by just dropping the table and creating a new one. Which was an okay thing to do, since the table only belonged to one project anyway. But I hate to think how long it would have taken to do a delete, cascading to child tables, if all the project data was in a single table. Since I'm not an expert in Postgres database design, I'm assuming I've done something sub-optimal. Are there some common techniques for tuning postgres performance? Do we need beefier hardware? Or is it a problem with how PHP or apache pools connections to the database?
