Re: looking for some real world performance numbers

Dave Cramer wrote:
> snacktime wrote:
>> I'm working through the architecture design for a new product. We
>> have a small group working on this. It's a web app that will be using
>> ruby on rails. The challenge I'm running into is that the latest
>> conventional wisdom seems to be that since obviously databases don't
>> scale on the web, you should just not use them at all. I have a group
>> of otherwise very bright people trying to convince me that a rdbms is
>> not a good place to store relational data because eventually it won't
>> scale. And of course we don't even have version 1 of our product out
>> of the door. I'll admit we do have a very good chance of actually
>> getting tons of traffic, but my position is to use a rdbms for
>> relational data, and then if and when it won't scale any more, deal
>> with it then.
>>
>> So what would really help me is some real world numbers on how
>> postgresql is doing in the wild under pressure. If anyone cares to
>> throw some out I would really appreciate it.

I missed the original post on this, so I'm replying to Dave's response.
To the OP, I don't know where you obtain your conventional wisdom
from, but I'd look for another source. Just about any site you might
visit that handles lots of data has a DBMS of some sort behind it; given
that IMS and Adabase have been out of favor for 20 years, most of those
DBMSs are relational. So if it can work for your bank, E*Trade and
eBay, chances are it can work for you.

As far as real world numbers, we have a data-intensive app (network data
collection for a telecom company) that is currently inserting about 16
million rows a day. I benchmarked PG for that app and with some
tweaking, PG could handle it. The current app uses stored procedures
for all inserts, and PG didn't do well with that approach; substituting
embedded inserts fixed that problem. So PG can definitely "handle" very
large transaction volumes. As with any DBMS and any application, you
may encounter challenges (like the one I point out with using stored
procs for high-volume inserts) that require you to address with some
thought.

--
Guy Rouillier