What architecture for a Large Database with PostgreSQL

[bmt]

I intend to set up a large database with PostgreSQL: about 1.3 TB, very basic queries (but 1500 per second at peak charge), and most probably only a single one index. The overall application, which is rather simple, should react in less than 3 seconds, 80% of the time. Meaning that if the database gives a mean response time of more than 0.5 second, I am fully dead...

I plan to set up a distributed cluster N+1. Each node of the cluster stores one independant part of the whole database on a SAN. A hash mecanism tells me which database to look for. The passive node

First, what cluster product (commercial or free) do you recommand on RedHat from your experience? In case of failover of the system, what down time should I expect until the passive node takes charge of the faulty node?

Second, what would be the reasonnable node size? If the cluster has 5 nodes, each node peaks up to 260 GB and 300 q/s. For each node, I expect to use biprocessors, upgradable to 4 CPUS in case of.
Does this seem reasonnable from your experience? or should I scale it down a bit? to 8 nodes +2?

Everyone around tells me I should go for Oracle. What do you think?

[Niheel]

The demands on the database seem to be high.

Would it not be easier to build on a commercial solution known to work in these types of situations such as Oracle. If you choose to go the open source route and this is your first venture into complex architecture; it would be good to have a consultant on hand that has done these things before with PostgreSQL. I am sure there are plenty of them supporting PostgreSQL.

Just thoughts on saving time and money.