Vertica Database

[Muller]

Hi Alf , I think you are one of the best to answer this. I don't remember where I first saw it, probably some article or ad but I became quite curious if they are saying works or not . Have you heard about them ? . Do you agree with them ? What about this column-oriented architecture. Do you think it's really better or it's just another market play ?

http://www.vertica.com/elqNow/elqRedir.htm?ref=http://www.vertica.com/_pdf/VerticaDatabaseDataSheet.pdf

Any real comparison with the relational model ?

tks..

[alf-admin]

[alf-admin]

Muller,

a further comment:

I suppose Vertica is a Sand type of organization. The storing of only columns allows a row to have only (associated) columns with values in it. That is what Sand.com does.

This is what I meant by the elimination of NULLs, a concept that has ridden the industry as a nightmare, because it has no meaning, but since a row (in physical storage) has a fixed set of columns, NULLs had to be allowed, from the vendor side, not the relational data model side. The relational data model states that tables are organized in rows with columns of arbitrary complexity that have values

Now that's something to chew on. Also types of arbitray complexity are included in the RM: What happended: We got integers, charachters (long charaters, dates. etc. Did we have a car type ? No.

But the car as a complex type (read: Object) is fully possible in the RM model. I wrote an opinion about it in this page:

http://www.databasedesign-resource.com/objects-vs-relational.html

Also about "flat" tables: An important one. More thoughts here:

http://www.databasedesign-resource.com/flat-tables.html.

Hoping I am somewhat clear here...
_________________
All the best,
Alf
Forum Admin
Light travels faster than sound: That's why so many people appear bright until you hear them speak...
www.databasedesign-resource.com
The Database Normalization eBook

[Muller]

Fantastic Alf , thank you for you reply. Your observations cleared a lot the things to , I was starting to think that E/R model was becoming obsolete. Now it seems more with my previous thoughts : There are 2 solid rocks where you can always build your path through computers "whenever" you are : Unix and Database.

tks..

[alf-admin]

You're welcome

Read more about NULLs here:

http://www.databasedesign-resource.com/null-values-in-a-database.html

(I know I have long URLs. SEs are finding my pages easier... Try this Google search for null values No. 1 out of 4.6 mil. )
_________________
All the best,
Alf
Forum Admin
Light travels faster than sound: That's why so many people appear bright until you hear them speak...
www.databasedesign-resource.com
The Database Normalization eBook

[andye]

Hi Alf,
This is a good thread (I work for Vertica). Your answer about handling nulls in column databases like Vertica was well described.

Regarding whether column databases really work or whether it's just a marketing play...I'm biased obviously. They really work, but the true test will be their adoption and usage over time.

2 proof points though, that might be of interest to people:

1-Don Feinberg, Gartner Group's analyst of the data warehouse database market talks about his views on column databases in a podcast on the Vertica web site - http://www.vertica.com/gartner. Gartner doesn't typically get excited about new technologies until they've seen lots of customers benefit from them. And he's pretty positive about column databases in the podcast ("outperform all other databases for analytic workloads").

2-Benchmarks - Vertica publishes customer benchmark results on its web site, and other column databases have published similar results--for the right workload (read/query-intensive) they provide much better performance than row-DBMSs (like Oracle, SQLServer, et al). But column databases aren't designed to handle OLTP workloads (write-intensive). Here are some benchmark results: www.vertica.com/benchmarks

I hope you found this helpful--I'm happy to provide additional views and welcome your feedback.

-ACE[/url]

[alf-admin]

Andy,

I highly appreciate your response, which I take is as a representative of www.vertica.com

Indeed, the trad. DB vendors have just followed the definition of the relational model, implementing a logical model (visualization) directly as a 1:1 physical mapping, without thinking "out-of-the-box" with regards to physical implementation. It's bound to go bad. And strictly column-oriented thinking is really not "out-of-the-box" thinking. Thanks for that.

When I read Muller's link about Vertica, it took me only seconds to recognize the new angle (thinking), and I was very pleased , having read similar thoughts from Sand a couple of years back.

[NotesTracker]

Vertica reminds me of another radical approach to DB architecture and implementation, which has been around for perhaps ten years or more.

This is Lazysoft's "Associative Model of Data" at http://www.lazysoft.com/technology_associativemodel.htm and their implementation of this called "Sentences Database" (see http://www.lazysoft.com/technology_sentences.htm )

Their initial implementation was solely on the IBM AS/400 platform -- these days called the IBM System i -- but I see from their web site http://www.lazysoft.com/ that today that Sentences is now available on other major hardware platforms and operating systems.

I see that you can download a PDF of Simon Williams' book "The Associative Model of Data" (which was once available in hard copy only), and a you can also download a free copy of Sentences Personal Edition. Not only that, but on their home page they state "Sentences Enterprise Edition is now available free of charge. Contact us for a copy." So there you are, something else for you to investigate and try out -- go for it!

Disclosure: I don't have any working or other relationship with Lazysoft at all. I used to be a database practitioner during my career at IBM, but these days am only a very interested observer of database technologies. Some people would call me a mere "dilettante" or "dabbler" but as an old player with a thick skin I'm not offended by this ...

I started at IBM Australia in 1970, so this is my 39th year in the industry! I saw the relational model was in its infancy, and watched it grow up to what it is now. When the first relational products (such as DB2) were first announced in the late 1970s and early 1980s, many DB pundits expressed strong doubts that relational DBs would ever perform adequately, compared with the then ruling network and hierarchical implementations such as IBM's IMS (which still operates at some organizations).

Of course, these days virtually all of the mainstream DBs are relational and the performance issues are well understood. It's very interesting, isn't it, to see the Vertica and Sentences style of products being conceived to fill gaps in the database product range, addressing perceived performance and/or functional shortfalls in the relational model.[/url]

[alf-admin]