Re: Larry Ellison and million bucks!

This is just the latest example of companies violating the spirit of the TPCD if not the actual letter of the law.

IBM DB2 has their covered index that the put up to 9 columns in a single index. At last count they were using 25+ indexes on the 9 tables. Not something that would be used in real world DW applications.

NCR Teradata responded with their Join Index. This put together a subtable that gives pointers to the rows to be joined. Speeds up joins but so slow to update that it has very limited application.

So Oracle now takes it one step further by actually materializing the views. I agree with you that they're pushing the envelope of what TPCD is about, but I don't think they're alone.

TPC is coming out with some new specs that should take effect next Spring. I'm hopeful that it will take away some of the maneuvering room that exists. However since the vendors make up the TPC, I'm not hopeful.

Bucko

Snorri Bergmann wrote:

> Here we go again.
>
> Remember when Oracle 7.1 was released? It set new TPC-C benchmark
> records all over the place. But when people started to read the full
> disclosure reports they saw a new 'feature' in Oracle 7.1 was being
> used: Discrete Transaction. DT was of no real value for customers and
> was basically a special feature that was coded in Oracle 7 specifically
> for the TPC-C benchmark.
> Oracle users: How many of you use the Discrete Transactions feature?
>
> Now Oracle has done it again!
> Have you heard about the features "Materialized views" and "Pre-computed
> aggregates" in Oracle 8i? These features allow Oracle to pre-compute the
> outcome of pre-defined queries during load!
>
> Now I read that "Oracle Corp. will pay $1 million to the first person --
> including anyone at Microsoft -- who can demonstrate that MS SQL Server
> 7.0 is not 100 times slower than the Oracle database when running a
> standard business query against a large database".
> More specifically, Oracle is talking about Query no. 5 in the TPC-D
> benchmark.
>
> I suggest people should not waste their time doing this benchmark
> because, as expected, Oracle is bending benchmark rules again (at least
> the spirit of the benchmark).
>
> Oracle claims: "Currently Oracle holds every database performance record
> in the world including data warehousing (TPC-D)..". Now, let's look at
> this statement.
>
> At www.tpc.org executive summaries of benchmarks can be obtained. Oracle
> has published two 1 TB benchmarks, both on Sun. Here are vital stats
> from these (An Informix 1 TB benchmark is included for comparison)
>
> Hardware DBMS Load time Time for
> Query 5
> Sun Ultra Enterpr. 10000 Oracle 8.04 27:02:02 1315.5 sec.
> Sun Starf. Enterpr. 10000 Oracle 8i 44:35:00 71.5 sec.
> Sun Enterprise 6000 Informix 8.21 11:55:23 561.5
> sec.
>
> Your first impression is probably: Oracle 8i is almost 20 times faster
> than Oracle 8.0.4!! However, load times are 50% longer, (in fact almost
> 400% longer than in Informix). This is because Oracle has pre-computed
> the outcome of all the TPC-D queries (filter, join, aggregate etc.)
> during load time and stored the results in internal tables. Great stuff
> for the TPC-D benchmark beacuse all queries are known in advance. So,
> Oracle 8i is probably reading 20 times less data in this particular case
> than 8.0.4.
>
> I can imagine that Data Warehouse administrators can't wait to get their
> hands on these new features because they know that
>
> 1) Load times don't matter
> 2) All queries in the DW are known in advance
> 3) DW queries are NOT ad hoc in nature
>
> NOT!
>
> Take care,
> -Snorri
> --
> Snorri Bergmann | Mail: snorri_at_strengur.is
> Strengur Inc. Armuli 7 | sbergman_at_informix.com
> 108 Reykjavik Iceland | Phone: +354 550 9000 (9007 direct)
> http://www.strengur.is/ | Telefax: +354 550 9010
Received on Wed Nov 25 1998 - 00:00:00 CST