1

I have 10 tables.

dbo.Table2008
dbo.Table2009
...
dbo.Table2018

Each table holds approximately 500,000,000 records which are 20 columns wide (if that matters). And each table has a clustered columnstore index on it.
Each table also holds data only for the year in its name. For instance, in dbo.Table2008 you'll only see records for which the CreatedDate column is >= 20080101 and <= 20081231. But if I want to issue a query across 3 years I need to hit all three tables with a union.

My theory is if I have a single table called dbo.Table which is partitioned into ~120 partitions, one partition for every month of every year, then it'll not only reduce my table count. So I can change my query from

with cte as (
    select col1, col2, col3, col4
    from table2008
    where col4 >= 20080201 and col4 <= 20120801
    union
    select col1, col2, col3, col4
    from table2009
    where col4 >= 20080201 and col4 <= 20120801
    union
    select col1, col2, col3, col4
    from table2010
    where col4 >= 20080201 and col4 <= 20120801
    union
    select col1, col2, col3, col4
    from table2011
    where col4 >= 20080201 and col4 <= 20120801
) select ...
from cte
join LookupTable1 on ... = ...
join LookupTable2 on ... = ...

to

select col1, col2, col3, col4, lookuptablecol1, lookuptable2col2
from dbo.Table
join ALL MY LOOKUP TABLES

Is my theory correct?
Is table partitioning with parition schemes and partition functions what I'm after?
And will this theory work using clustered columnstore indexes?

  • First, will they improve performance. Second, I'm sure they'll clean up my queries against the table by eliminating the CTE. Third, will I be able to keep a clustered columnstore index on this table? – gh0st Apr 26 '18 at 14:29
1

Partitioning does not sound like a good fit for your scenario. Partitioning isn't something to be taken lightly. It requires a lot of planning and you'd have to rewrite 5 billion rows in order to convert the data. Instead you should consider a UNION ALL view. Depending on your requirements you could even make it a partitioned view.

A view will give you an easy way to query your data without any date movement. It's by far the simplest solution.

In general, partitioning doesn't improve query performance. However, with columnstore tables it can be helpful if it gives you better rowgroup elimination on the partitioned column. That can be accomplished without partitioning depending on how you build your columnstore tables. If you don't modify the old data it should be relatively straightforward to build your columnstore in a way that maintains order.

Columnstore tables do support partitioning, if you do have a legitimate need for it at some point. It's easy to experiment with it. Simply try to create an empty, partitioned columnstore table.

0

I am using the same partitioned tables in my company like you mentioned, but only difference is, I have monthly table and not yearly. And each table volume is around 50 GB.

For querying the tables which belongs to different months, I use dynamic sql to select correct table based on the value I pass to the variable. If I pass a year then my query will join all the tables belongs to that year and looks for the data. All this logic will be from the proc. So, I just add parameters to the proc to build the dynamic sql.

Example:

DECLARE @YYYYMM INT = 200801
DECLARE @SearchString VARCHAR(150) = 'customer1'
DECLARE @TSQL VARCHAR(5000)
DECLARE @Table VARCHAR(100)

SET @Table = 'Table' + CONVERT(VARCHAR,@YYYYMM)
SET @TSQL = 'SELECT * FROM ' + @Table + ' WHERE Column1 LIKE ''%' + @SearchString +'%'''

EXEC (@TSQL)
  • I understand the use of dynamic sql. This is the scenario I'm using right now but I have a hunch that partition functions/partition schemes are supposed to help with this. I'm trying to confirm my hunch. – gh0st Apr 26 '18 at 14:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.