Forum OpenACS Development: Re: Scalability of acs_objects and acs_object_context_index

Let's put it this way ... it's a LOT faster than it, and it doesn't degrade spectacularly in performance when you add lots of parties, objects, etc the way it used to.

Whether or not it is fast enough for a particular real-life site is an open question, though.  Testing is advised!

Also I'd still like to flatten out the number of defined permissions.  If we didn't have 20 different "foo_read" privileges performance would be better.  Currently the more packages you add the poorer the permissions system performs simply because of all these "foo_read", "foo_write" and "foo_admin" permissions all of which are children of the global "read", "write" and "admin" perms which adds to the denormalized hierarchy maps.

Server prices are getting awfully low for impressive performance, though, my new SCSI based Dell 600C, which after a second disk (for software RAID 1) and an extra 512 MB registered ECC RAM came in at about $800, greatly outperforms the dual P450 SCSI based system I built for $2100 four years ago and has 8 times as much RAID 1 storage and the disks are 1/3 faster with huge 8MB buffers, too.

Not that this means we want to wantonly burn cycles, but the flexibility of the permissions system certainly is attractive and high throughput servers are cheaper and cheaper ...