Forum OpenACS Q&A: RDBMS clustering or replication

Collapse
Posted by Andrew Piskorski on
Ludek, AFAICT a cluster of multiple quad CPU machines is huge, massive overkill for all but a minuscule fraction of RDBMS backed websites, even if you are also integrating substantial amounts of data that is not directly web-related. (One modern dual CPU box with 16 or 32 GB of RAM can do a lot...) So you aren't likely to find all that much real interest in such RDBMS clustering in the OpenACS community.

Do you seriously anticipate a need for such clustering at AIESEC, or are you just asking for background knowledge? What does your current RDBMS dataset, usage, hardware, and load look like?

I have not tried out Oracle 10g's clustering features at all yet. I will sometime, but for its remote offsite backup and failover capabilities, not in an attempt to increase performance. My (limited) understanding is that this is what all of Oracle's "clustering" functionality is primarily aimed at - data integrity and business continuity, not increased throughput. In fact I believe using Oracle clustering tends to slow down your RDBMS slightly vs. the single box case, not speed it up. (You may, of course, get a speed up if you use the slave nodes for large slow read-only queries, thus unloading those from the master node.)

PostgreSQL does have extensive replication and failover capabilities now, but I'm not particularly familiar with them, nor how they compare to what Oracle 10g has. That said, I think PostgreSQL does not yet have multi-master (writes allowed on multiple nodes) capability (Oracle does), but they're working on it.

For some related links see these older threads: one, two, three.