Re: ΄πΈ΄: Horizontally scaling a database

Many years ago I worked with an in-core database that was limited by the size of the (then available) RAM.

We got around this by using multiple, identical (except for names) databases. Customers A-L used database (for your purposes, a table) ONE while customers M-Z used TWO. Actaully, there were thousands but the principle holds.

To query the entire data store we used views that did unions across database links.

Performance will be terrible, but it does what you asked for.

Go out and get a bunch of Pentium III machinces surplus with 40GB hard drives. Load Oracle and create the identical database on each machine (just vary the SID). Put them all on a network and have a master machine whose database is nothing but views and database links.

Have your customers just work with their "local" database. You can probably add some business smarts and put in local views that will link just to the other databases that each customer needs.

Backing them all up is a pain; however, the whole system is remarkably robust since a single machine or database failure will only affect a limited number of customers.

Have fun!

--
http://www.freelists.org/webpage/oracle-l