By using this site, you agree to our updated Privacy Policy and our Terms of Use. Manage your Cookies Settings.
454,278 Members | 1,502 Online
Bytes IT Community
+ Ask a Question
Need help? Post your question and get tips & solutions from a community of 454,278 IT Pros & Developers. It's quick & easy.

about multiple fact tables and star schema

P: n/a
HI,

I had a special senario here..

I need to build a database which only around 50-70GB finally.

But one of table would be 40GB finally. ( grow every year until to the
retension period)

rest of the tables are very small . (MB size level)

I'm thinking put the database on one partition, but had some query
performance concern on
that big table.

I'm going to break down the table by the time. But had one concern. It
is data warehouse environoment, Currently, this big table is a fact
table. it is a star-schema model. If I build a view based on the break
down tables, is that any impact on the optimizer to use star schema
model to choose access plan?

Aug 21 '06 #1
Share this Question
Share on Google+
6 Replies


P: n/a
Jane,

Which version of DB2, which platform?
In DB2 9 you can easily use range partitioning.
In both DB2 V8 and DB2 multi dimensional clustering may apply.

UNION ALL view would be my last choice.
Even 4KB Pages give you up to 64GB on DB2 V8.
How much headroom do you need?

Cheers
Serge
--
Serge Rielau
DB2 Solutions Development
IBM Toronto Lab

IOD Conference
http://www.ibm.com/software/data/ond...ness/conf2006/
Aug 21 '06 #2

P: n/a

Serge Rielau wrote:
Jane,

Which version of DB2, which platform?
In DB2 9 you can easily use range partitioning.
In both DB2 V8 and DB2 multi dimensional clustering may apply.

UNION ALL view would be my last choice.
Even 4KB Pages give you up to 64GB on DB2 V8.
How much headroom do you need?

Cheers
Serge
--
Serge Rielau
DB2 Solutions Development
IBM Toronto Lab

IOD Conference
http://www.ibm.com/software/data/ond...ness/conf2006/
Thanks, Serge.

Probably I did not explain my question clearly.

My environment is UDB V8 on AIX.

My concern is not table size limit. Because now I know the Max size
for the big table is 40GB. It is OK on single partition, even for 4K
page.

My concern is on query performance. I know if using multi-partition, it
must be resolved.
but the other tables are all very small, and the total DB size is not
big (50GB), based on the DB size, I want to use single partition
database. Only because of this big table to use multi-partiton , seems
a little bit waste..

That's why I want to break down the big table to relatively smaller
ones. Use union all view.

One of my colleague remind me this would change optimizer to not use
star schema to get good access plan..
(because currently , the big table is fact table, it is in star schema
model)
I'm not sure if this is the case..

Aug 21 '06 #3

P: n/a
Make your fact table a multi-dimensional-cluster.
That's ll do. I agree that DPF would be overkill.
If you don't have a low cardinality column that offers itself up for
MDC, just add another column e.g.
yearmonth GENERATED ALWAYS AS (INTEGER(date)/100)
Then cluster on the yearmonth column.
If you have 3 years worth of data this would partition the table into
3 * 12 =36 slices.

Cheers
Serge

--
Serge Rielau
DB2 Solutions Development
IBM Toronto Lab

IOD Conference
http://www.ibm.com/software/data/ond...ness/conf2006/
Aug 21 '06 #4

P: n/a

Serge Rielau wrote:
Make your fact table a multi-dimensional-cluster.
That's ll do. I agree that DPF would be overkill.
If you don't have a low cardinality column that offers itself up for
MDC, just add another column e.g.
yearmonth GENERATED ALWAYS AS (INTEGER(date)/100)
Then cluster on the yearmonth column.
If you have 3 years worth of data this would partition the table into
3 * 12 =36 slices.

Cheers
Serge

--
Serge Rielau
DB2 Solutions Development
IBM Toronto Lab

IOD Conference
http://www.ibm.com/software/data/ond...ness/conf2006/
Serge

Thanks a lot!

I did not realize I could use MDC...

Aug 22 '06 #5

P: n/a
Serge,

Your response brings up a question. How will the advantages of MDC
tables be leveraged when the clustering column is not related to the
actual data relationships and therefore will not be referenced in any
predicates?

Thanks.

Lew

Serge Rielau wrote:
Make your fact table a multi-dimensional-cluster.
That's ll do. I agree that DPF would be overkill.
If you don't have a low cardinality column that offers itself up for
MDC, just add another column e.g.
yearmonth GENERATED ALWAYS AS (INTEGER(date)/100)
Then cluster on the yearmonth column.
If you have 3 years worth of data this would partition the table into
3 * 12 =36 slices.

Cheers
Serge

--
Serge Rielau
DB2 Solutions Development
IBM Toronto Lab

IOD Conference
http://www.ibm.com/software/data/ond...ness/conf2006/
Aug 22 '06 #6

P: n/a
se*****@yahoo.com wrote:
Your response brings up a question. How will the advantages of MDC
tables be leveraged when the clustering column is not related to the
actual data relationships and therefore will not be referenced in any
predicates?
It won't. That's the same for UNION ALL, or range clustering.
If there is no way for the compiler or the run time engine to eliminate
ranges then its' useless.
If you are looking at parallelism note that SMP doesn't operate on a per
range level. Instead it will start reading all the ranges in all the
threads and sub divide within the range.
This is different from DPF where each node will read it's database
partition exclusively.

Cheers
Serge
--
Serge Rielau
DB2 Solutions Development
IBM Toronto Lab

IOD Conference
http://www.ibm.com/software/data/ond...ness/conf2006/
Aug 22 '06 #7

This discussion thread is closed

Replies have been disabled for this discussion.