Re: Need advice !!

damorgan <dan.morgan_at_ci.seattle.wa.us> wrote in message news:3C5EC972.3CC91E93_at_ci.seattle.wa.us...
> I won't flame you give the caveat that only 4 disks are available. But it
> certainly will not be a good performer.
>

Define "good".

We seem to be having a case of what I'm going to call "ferrari syndrome" here. A ferrari is a good performing car, but so is a ford fiesta if all you're doing is nipping to the shops.

I have a data warehouse (more like a shed, it's only 3GB) on five (old) 18GB drives in RAID 5 on a twin 500MHz pentium II with 1 GB of RAM.

It's rebuilt in its entirety every night in two stages. The first takes under 15 minutes. The second takes 30 minutes. If I could be bothered, I could probably reduce the each stage to < 10 minutes.

Interactive queries on the most detailed aggregate table (about 1/10 of the size of the raw data) generally take less than 1/3 of a second. Should that be an issue, I can use materialised views to reduce that to 1/50th of a second. Similar queries are used to drill down until at a given level of detail, the source transactions can be selected, generally taking < 10 seconds to display.

This is, I think, enough of a "good performer" to be more than acceptable

Even if there was 10x the amount of data (ie 30GB), I believe I could maintain similar interactive query times and get the overnight update to run in < 4 hours. The issue actually becomes one of concurrent usage and effective caching.

If there was 300GB of data then I would obviously have to think again, but then I would also have to have more disks in any case. We know the original poster has a datawarehouse that fits in less than 72GB.

Also, I would take issue with your comment about redo logs not being RAID5. Given good coding, the load on redo logs for a datawarehouse should be negligible, so it doesn't matter where they go.

    Argumentatively yours

        Keith Received on Mon Feb 04 2002 - 14:56:07 CST