I’d missed that Teradata feature. Thanks!

DATAllegro does “multi-temperature” in another way. In essence, they let you put two different brands of disk in the same installation, one of which is cheaper/bigger/slower than the other. Then they manage it all as a fairly seamless logical whole, as part of their grid strategy.

CAM

One nice feature that Teradata now offer is the ability to mix “hot” (frequently accessed) and “cold” (rarely accessed) data on the same disc spindles.
This gets around the limitations of disk reader speeds vs increased density discs.
If say only 20% of the data on a spindle is hot then it stands to reason that the data can be read from many spindles that bit quicker than trying to read all the hot data from fewer spindles.
Maybe other products have similar functionality…
BTW I currently work with SQL Server so I am not trying to promote TD ! Too expensive IMHO.

John

I haven’t talked enough with Dataupia to make many en passant comments about them.

E.g., I just caught an error in the bullet you commented on — I had implied ParAccel’s Oracle compatibility was already available, when in fact that’s a “roadmap” item, with a vague delivery timeframe target that they freely admit isn’t set in stone. I knew enough to edit that, without hesitation. For Dataupia I wouldn’t be as sure about what’s real and what isn’t.

CAM

I’ve been an analyst since 1981, and I’ve always dealt with huge amounts of semi-confidential and only partly-verified info. (My stock analyst years were at the beginning of that period, long before Reg FD.) A big part of an analyst’s skill lies in figuring out what’s actually for real.

I’m usually right. Not always, to be sure. But usually. At least about reading the current state of the market and technology. Predictions are sometimes a different matter …

Best,

CAM

Cheers.

http://www.dbms2.com/2006/09/28/expansion-ratio-data-warehouse-explosion/

As for naming the big databases — I’m sorry. Most of the info I get about particular users is under possible NDA, and I often don’t bother checking what has or hasn’t made it into press releases and the like. However, I do know DATAllegro has a press release about a multihundred terabyte site. Dataupia put out a release about a big user very recently. (If memory serves, they called it 150 terabytes, but upon query from me said it was 120 Tb of user data. I haven’t posted about it yet.) eBay, which I just posted about, is not — I’m pretty sure — running its main warehouses on Oracle or SQL Server. If I get clarity for public consumption, I’ll update accordingly. Walmart has huge data warehouses, and is a long-time Teradata customer. Ditto K-Mart pre-Sears-merger. I posted about a 35 Tb Sun/Greenplum site where Oracle had been thrown out. And so on.

If you really want to learn and not just argue, there are a lot of pointers out there. If nothing else, just take the obvious suspects for the largest warehouses, leave out whoever you think will be bewildering due to decentralized IT decision making or secrecy (hint: start that list with Amazon and the intelligence community), and start seeing who’s known to be their vendor.

Have you even read my prior posts on these subjects? There’s a reasonably decent search capability on my sites.

CAM

WRT “unnatural acts”: I’ve never had to do this, nor heard of anyone who has, and would not see why it would need to be done. Seems to me like a case of poor SQL writing skills rather than an Oracle product limitation. If you have a test case I’d gladly look at it offline.

Can you give the definition that you use for “user data” and explain the difference between the Winter Corp definitions?
Why is this number 1/5 - 1/10 as much in Oracle data warehouses? Is this ratio also applied to other data warehouse platforms?

CAM

As for “unnatural acts” — can you write the SQL one would think you can write, or do you have to union over the result sets from a partitioned set of tables to get your results?