An interesting interview indeed.

I am yet to compare what Mike Stonebraker just said, with corressponding points from Oracle.
Let me search oracle’s answer to this.

A few years ago Ant-DB claimed it was 3 times faster than Oracle (leave aside sybase, SQL server etc), Around 4 years have passed snice I heard that and I am yet to see their installations around.

CUIL also says they can overapower google.

They say, Reality is stranger than fiction.
Let’s wait and watch.

At the same time, I have full confidence in Mike Stonebraker’s capabilities too (having seen his last 25 yrs career track).

Cheers!

Param
send2param@yahoo.com

http://www.databasecolumn.com/2008/07/debunking-another-myth-columns.html

that debunks this myth.

Daniel Abadi
Assistant Professor
Yale University

Mike Stonebraker’s point that it is more efficient to query a table of 3 commonly used columns instead of projecting them from a table of all 200 columns is undeniably true, however as Oracle practitioners we do not have to accept a scan of a 200 column table as our only choice. We have the tools at hand to define skinny tables and aggregate tables (and combinations of both using HOLAP MV’s or OLAP MVs) that allow the optimizer to choose which one to query. The tables compress better, of course, and for single columns we can also provision extremely fast bitmap indexes.

So while Vertica tackles this particular problem, Oracle has its own solution for exactly that same problem.

I recall about 6 years ago that the business management at the organisation I was working for approached us, the technical team, with a proposition from a company that provided an indexing middle-tier between applications and Oracle data warehouses. Their major selling point was the acceleration of queries based on single columns — “count 60 million rows in two seconds!” … “add 100 million rows in 20 seconds!” … that sort of thing. In every case we were able to provide comparisons with real world performance benchmarks from our Oracle 9i system based simply on bitmap indexes or other common techniques.

I’m afraid that I see Vertica as something of a similar one-trick pony, and I haven’t seen any advertised capabilities that would not be more simply implemented within Oracle. I wonder whether in any of the cases that are quoted in favour of Vertica’s performance gains, were any attempts made to use Oracle to match those improvements by using these techniques?

The column storage is a great idea but I don’t have until now the chance to try it.

In the same domain, you have also the “vertical partitionning” which is a solution for the same problem (more than 400 column in the fact table for only a few used).
http://asktom.oracle.com/pls/asktom/f?p=100:11:0::::P11_QUESTION_ID:1277688700346608142

Happy Christmas

We see from the widespread use of Microsoft, and Oracle that standardisation is a secret to success. It does not have to be the best solution, there are other factors which come into play – Support, development costs, skills base, ease of use, etc.
Oracle have been giving away the software for free, to help developers understand it, and there is a huge skills base, with mountains of books to read – How does a new db on the block compete with this?

Nice one !