Some clarifications
Let me first address an error in the article. When describing the "skipping" and the "columnar properties" of BLU the example is reversed. Sales would benthe column and the sales yerasmallow formthe skipping. One wouldn't want to have 10 columns for each year.
Now, to the comments:
What's available for Informix and DB2 is an accelerator. That's a separate add-on to which IDS and DB2 z can reach out. The name of that project was "Blink". BLU stands for Blink ULtra. Think of it as second generation.
It is deeply integrated and uses some new additional features Blink does not.
Now, comparing to Sybase IQ.
Yes Sybase IQ is columnar, and yes Sybase IQ uses compresion.
But BLU goes beyond that. The fact that DB2 can perform comparison functions on the compressed data gives a big boost. There is no need to uncompress first, and then evaluate predicates.
This is what Netezza does in FPGA, DB2 z with hardware support and Oracle and DB2 with deep Compression in software.
The SIMD vector processing here is also novel here, I am not aware of Sybase IQ or Oracle exadata doing hat.
Skipping is not a WHERE clause. Or rather the SQL is a where clause, but BLU has technology loosely borrowed from Netezza which allows it to bypass large areas of the table altogether, similar to what you would do with partitioning, but without the overhead of administration.
Finally what makes BLU exciting is that it is within a regullar DB2. You can use BLU technology in an operational warehouse side by side with transactional operations including joins across the reporting (BLU) and transactional (regular) tables.. Sybase IQ is suitable only for datamarts since pure column stores cannot handle transactions well.
All, in all, it's not any of the individual technologies that make BLU great. It is the sum of them, and the fact that it's integrated into a mature DBMS with all the bells and whistles.
Cheers
Serge
Chief Architect IBM OpenPages