Hi Stefan, Thank you for the information, now I see the point.

Did you by any chance try any other DBMSs with your "non-standard" workload? If so, how do those experiences and limitations compare to MonetDB?

I did not run tests over conventional DBMSs because they have stricter limits compared to MonetDB (e.g. Postgres is limited to 1600 columns), but only over array DBMSs or math packages. Obviously, their column scalability is better, but they can not handle RA that efficiently. Best regards, Oksana On Tue, 16 Oct 2018 at 12:16, wrote:

Send users-list mailing list submissions to users-list@monetdb.org

To subscribe or unsubscribe via the World Wide Web, visit https://www.monetdb.org/mailman/listinfo/users-list or, via email, send a message with subject or body 'help' to users-list-request@monetdb.org

You can reach the person managing the list at users-list-owner@monetdb.org

When replying, please edit your Subject line so it is more specific than "Re: Contents of users-list digest..."

Today's Topics:

1. Re: Column limit (Oxana Dolmatova) 2. Re: Column limit (Stefan Manegold)

----------------------------------------------------------------------

Message: 1 Date: Mon, 15 Oct 2018 17:14:14 +0200 From: Oxana Dolmatova To: E.Rozenberg@cwi.nl Cc: users-list@monetdb.org Subject: Re: Column limit Message-ID: Content-Type: text/plain; charset="utf-8"

Hi Eyal,

Sorry for the late reply. Yes, there is no problem with large number of columns in sparse representation of arrays (i.e., dimension, dimension, value). However, in order to perform relational operations, this structure should be pivoted first. When matrices are represented by tables, both algebras can be applied to the same object, so I'm specifically looking for the main bottleneck of processing hundreds of thousands of columns in MonetDB.

Best regards, Oksana

On Wed, 10 Oct 2018 at 13:40, Eyal Rozenberg wrote:

...

If you want to model an NxM matrix of elements of type MYTYPE in a DBMS (not just MonetDB), you would typically have a table with 3 columns:

CREATE TABLE my_matrix ( "row" INTEGER NOT NULL, "col" INTEGER NOT NULL, "cell_value" MYTYPE NOT NULL, PRIMARY KEY ("row", "col") );

of length N x M.

Anyways, I'm not the expert on Arrays in our group so I'll let others chime in.

Eyal

...

Hello,

Thank you for all the replies and information! The version is 11.23, plus I'm using the cloud platform with 4cpu-16ram-hpc. The number of rows was just 1K for 15K attributes in a table. I understand that MonetDB was made with no intention to create and process tables with that many columns. But I'm interested in matrix operations integration, so I would like to know what is the main bottleneck to process such data (matrix-like sized) in the system.

Best regards, Oksana

On Tue, 9 Oct 2018 at 21:02, Eyal Rozenberg mailto:eyalroz@technion.ac.il> wrote:

I'll add two observations to what Zhang Ying said.

1. Internally in MonetDB, the index of a column (or a BAT) in MonetDB's BBP (BAT Buffer Pool) is a signed 32-bit integer (on most

On 10/10/2018 13:01, Oxana Dolmatova wrote: platforms), so

...

there isn't a theoretical problem having more than 15,000 columns;

but

...

MonetDB was not designed with that many tables in mind. The "long dimension" is expected to be the length of columns.

2. When you have as many as 15,000 columns, you're not simply

finding

...

more attributes of the same items. So, for example, a person has a

first

...

name, last name, year of birth, etc. etc. - but after several

dozens of

...

these you're probably adding columns such as "data relevant to this person at point X", "data relevant to this person at point Y" and

so on.

...

In your case, I'm guessing you have a lot of points and hence a lot

of

...

columns with data for your different items. What you should

consider

...

doing instead is having an extra "coordinate" column. Thus, in the example I gave, I would drop most of the columns, in favor of a

second

...

table named "person_data". This table's columns would be

"person_id" (or

...

some other key columns), "sample_point" and "value". Thus instead

of M

...

columns and N records you would have N x M records with 3 columns.

Eyal

On 09/10/2018 16:50, Ying Zhang wrote: > Hai Oksana, > > A.f.a.I.k., there is no explicit limit on the #columns you can

put

...

in a relation (did you mean ?table??). The reason that mserver5 stopped responding is probably because it is busy with I/O. > > A long time ago, I once tried to create 1M tables, each with 3 columns. After about a day or something, MonetDB has created ~400K tables, and I could hear my HDD being constantly busy. MonetDB was still working, but that was already a sufficient evidence that that idea wasn?t going to work. > > So the question is actually what do you want to achieve? > > Regards, > Jennie > > >> On 9 Oct 2018, at 13:43, Oxana Dolmatova mailto:oxana.dolmatova@gmail.com>

wrote:

...

>> >> Hello, >> >> Could you tell the limitation for number of columns in a

relation

...

and the main reason for that limitation. So far I could create relations with up to 15000 columns. When the number of columns is bigger mserver5 stops responding. >> >> Thank you and best regards, >> Oksana Dolmatova >> _______________________________________________ >> users-list mailing list >> users-list@monetdb.org mailto:users-list@monetdb.org >> https://www.monetdb.org/mailman/listinfo/users-list > > _______________________________________________ > users-list mailing list > users-list@monetdb.org mailto:users-list@monetdb.org > https://www.monetdb.org/mailman/listinfo/users-list >

-- Best regards, Oksana Dolmatova

_______________________________________________ users-list mailing list users-list@monetdb.org https://www.monetdb.org/mailman/listinfo/users-list

-- Best regards, Oksana Dolmatova