partitioning a table question...

Hi,
I have to plan a datawarehouse structure where the primary reporting usage
is to list detailed information about customers.
And also, we have OLAP cube for analysis purpose.
I'll have 100 000 customers and a fact table containing between 35 to
45millions of rows by year. (we also have 2 other fact table with 5 to
6millions of rows by year)
Generally, this fact table is filtered for a particular organizational unit
and for data in 1 year. (for report generation using report server)
So, I plan to partition my fact table by organizational unit and by year.
But I have 180 units.
Does the partitioning will works fine with 180 * 5 years = 900 tables?
on this article:
http://msdn.microsoft.com/library/de...itionsindw.htm
there is a note that the maximum number of tables is 256.
But the performance gain can be very high!
The estimated size of the DW is 10gb / year (maybe more with additional
indexes)
My DTS package is ready to support partioning table loading. (the package
automatically create the new table, indexes and update the view for each
detected partition in the staging source table)
My users access my reportserver interactively, scheduling anything is not an
option.
For higher analysis, my olap cubes are ready.
thanks for your feedback.
Jerome.
Hi Jerome:
I have some kind of "rule". Just make a partition for each 1 GB of data.
Or doing it for a year, would be great. If you need more granularity, give
it a try to month partitioning.
Tell me if you need some help on this
Sincerely
Alejandro Leguizamo
MVP SQL Server
Colombia
"Jj" <willgart@.BBBhotmailAAA.com> wrote in message
news:uwuxtzxzEHA.3336@.TK2MSFTNGP11.phx.gbl...
> Hi,
> I have to plan a datawarehouse structure where the primary reporting usage
> is to list detailed information about customers.
> And also, we have OLAP cube for analysis purpose.
> I'll have 100 000 customers and a fact table containing between 35 to
> 45millions of rows by year. (we also have 2 other fact table with 5 to
> 6millions of rows by year)
> Generally, this fact table is filtered for a particular organizational
> unit and for data in 1 year. (for report generation using report server)
> So, I plan to partition my fact table by organizational unit and by year.
> But I have 180 units.
> Does the partitioning will works fine with 180 * 5 years = 900 tables?
> on this article:
> http://msdn.microsoft.com/library/de...itionsindw.htm
> there is a note that the maximum number of tables is 256.
> But the performance gain can be very high!
> The estimated size of the DW is 10gb / year (maybe more with additional
> indexes)
> My DTS package is ready to support partioning table loading. (the package
> automatically create the new table, indexes and update the view for each
> detected partition in the staging source table)
> My users access my reportserver interactively, scheduling anything is not
> an option.
> For higher analysis, my olap cubes are ready.
> thanks for your feedback.
> Jerome.
>
>
|||but does SQL Server support 500 tables in a partitioned view?
I don't need help about this, I just want to plan correctly.
and I don't want month partition but only year + organizational unit
Because I don't do sum or count aggregation in SQL (or just a little sum
compared of what my cube provide).
Generally my reports contains SQL statement like this:
* Last kown value at a specific date (subquery required) (by customer)
* events which start before and end after a specific date (between
statement) (by customer)
Its the reason of not using monthly partitions, because I never know when
these dates appear in the database.
"Alejo Leguizamo (MVP SQL)" <SQL@.sql.sql> a crit dans le message de news:
%23thX4VC0EHA.1204@.TK2MSFTNGP10.phx.gbl...
> Hi Jerome:
> I have some kind of "rule". Just make a partition for each 1 GB of data.
> Or doing it for a year, would be great. If you need more granularity, give
> it a try to month partitioning.
> Tell me if you need some help on this
> Sincerely
>
> --
> Alejandro Leguizamo
> MVP SQL Server
> Colombia
> "Jj" <willgart@.BBBhotmailAAA.com> wrote in message
> news:uwuxtzxzEHA.3336@.TK2MSFTNGP11.phx.gbl...
>
|||Jerome,
Views are based on select statements. A select statement can have a max of
256 tables in the From clause. So the answer to your specific question is
no.
Danny
"Jj" <willgart@.BBBhotmailAAA.com> wrote in message
news:e1ghnJD0EHA.2624@.TK2MSFTNGP11.phx.gbl...
> but does SQL Server support 500 tables in a partitioned view?
> I don't need help about this, I just want to plan correctly.
> and I don't want month partition but only year + organizational unit
> Because I don't do sum or count aggregation in SQL (or just a little sum
> compared of what my cube provide).
> Generally my reports contains SQL statement like this:
> * Last kown value at a specific date (subquery required) (by customer)
> * events which start before and end after a specific date (between
> statement) (by customer)
> Its the reason of not using monthly partitions, because I never know when
> these dates appear in the database.
> "Alejo Leguizamo (MVP SQL)" <SQL@.sql.sql> a crit dans le message de news:
> %23thX4VC0EHA.1204@.TK2MSFTNGP10.phx.gbl...
>
|||
sniff
I think I'll do this: partition by organization for the current year and by
year for the history... does this works in a partitioned view?
"Danny" <istdrs@.flash.net> a crit dans le message de news:
7Lkod.23022$Rf1.15232@.newssvr19.news.prodigy.com.. .
> Jerome,
> Views are based on select statements. A select statement can have a max
> of 256 tables in the From clause. So the answer to your specific question
> is no.
> Danny
> "Jj" <willgart@.BBBhotmailAAA.com> wrote in message
> news:e1ghnJD0EHA.2624@.TK2MSFTNGP11.phx.gbl...
>
|||The primary key of the underlying tables in the view must be the same. For
this to work, you will have to query current and history separately with
each haing its own view.
Danny
"Jj" <willgart@.BBBhotmailAAA.com> wrote in message
news:uKoQZzP0EHA.1932@.TK2MSFTNGP09.phx.gbl...
>
> sniff
> I think I'll do this: partition by organization for the current year and
> by year for the history... does this works in a partitioned view?
> "Danny" <istdrs@.flash.net> a crit dans le message de news:
> 7Lkod.23022$Rf1.15232@.newssvr19.news.prodigy.com.. .
>