Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: Large table and buffer cache concerns
Thank you both for your feedback. To add additional clarity:
> The two important questions I couldn't quite get from your
> posts: will you be querying a large proportion of table j at times
> (3000 customers all coming on at 9AM?), and what kind of _updates_
> will you be doing on table j with multiple users looking at the same
> blocks?
While not being at liberty to get into too many details, I'll use the example of Netflix's movie list.
Let's assume:
Netflix only has 400K customers. (table c)
Netflix only rents 1000 movies. (table q)
Any of Netflixes customers can start up a web browser and view the
list of movies they have in their queue and add new movies to their
queue at any time. Also these same customers can rate the movies
they've viewed. (table j)
A single web server for Netflix is capped at 3K customers it can
support at one time.
At any given point in time there are always 3000 customers logged in
changing their movie queue and rating the films they have seen. But
it could be any 3000 of the total 400K customers Netflix has.
This next part defies logic but I don't want to discuss the details of why we reload the customers list once very 5 to 30 min, let's just assume we have a really good reason for doing so. On average, once every 5 to 30 min per online customer (not all at the exact precise time, a random interval), we reload their movie list (select up to 1000 records).
As to update frequency, let's supsend reality in our example and allow a customer to add a new movie to their list once every 30 min (insert), 5 min later we ship the movie to the customer (update), 30 min later the customer has viewed the movie and has returned it (update). 5 min after that they review/rate the movie (update). Rinse repeat, up to three movies at a time per customer. We never delete the list of movies the customer has queued, viewed and rated.
I'm trying really hard to beleive in normalization and the relational way. Received on Wed Apr 11 2007 - 14:49:54 CDT