# Improving Cube Performance with Precalculated Aggregations

• Comments posted to this topic are about the item Improving Cube Performance with Precalculated Aggregations

• Very cool article, I utilize precalculation all of the time, but never had to do closing balances or similar to that, but definitely great to know. The SQL to calculate closing balances seems more complicated than I would have expected, but if it does the job quickly and efficiently than it works.

Rob

• Here's a much simpler way to calculating the CloseBalance. No recursion or temp tables. A temp table for the daily total would definitely be faster especially in future dates, just keep adding to it daily since the GL should be immutable.

If validated the SQL but haven't tested it against their DB. I based it off one for order totals in a commerce app and worked great.

`WITH tvDailyTotal AS`

`(`

`SELECT Company_Key, Account_Key, SUM(ISNULL(Amount,0)) DailyTotal, [Date]`

`FROM Fact_GL`

`GROUP BY [Date]`

`HAVING DailyTotal != 0`

`)`

`SELECT Company_Key, Account_Key, TD2.[Date], TD2.DailyTotal, SUM(TD1.DailyTotal) Amount_CB`

`FROM`

`tvDailyTotal TD1`

`RIGHT JOIN tvDailyTotal TD2 ON TD1.[Date] <= TD2.[Date]`

`GROUP BY TD2.[Date], TD2.DailyTotal`

`ORDER BY TD2.[Date]`

• Hi Rob,

Thanks for your feedback! Sure you can write the closing balances calculations in much shorter ways (like your example), but performance would be severely reduced. The purpose of the recursive query is not to complicate things ;-), but to reuse previous closing balances so the underlying rows are not needed to be read again.

Calculating closing balance for day n.

Plain Query: CBn = day1 + day2 + ... + dayn (n reads)

Recursive Query: CBn = CBn-1 + dayn (2 reads)

Let's consider an example where you have data for 1000 days. The plain query would then do about 500 000 reads, while the recursive query would do about 2 000.

Actually my query can be further optimized quite a bit, but I chose to go for readibility rather than absolutely maximum performance.

/Johan

• JAhlen (11/4/2009)

Hi Rob,

Thanks for your feedback! Sure you can write the closing balances calculations in much shorter ways (like your example), but performance would be severely reduced. The purpose of the recursive query is not to complicate things ;-), but to reuse previous closing balances so the underlying rows are not needed to be read again.

Calculating closing balance for day n.

Plain Query: CBn = day1 + day2 + ... + dayn (n reads)

Recursive Query: CBn = CBn-1 + dayn (2 reads)

Let's consider an example where you have data for 1000 days. The plain query would then do about 500 000 reads, while the recursive query would do about 2 000.

Actually my query can be further optimized quite a bit, but I chose to go for readibility rather than absolutely maximum performance.

/Johan

Very cool, I like the recursion option and definitely more efficient. I'll have to try it out!

• For the running total problem you may want to check this thread http://forum.lessthandot.com/viewtopic.php?f=17&t=7601&st=0&sk=t&sd=a&hilit=running+total

Interesting to see Moshe Pasumansky mentioned in this article. We studied in the same faculty (school) different departments in my youth (St. Petersburg Technical University).

• nnosonovsky (11/4/2009)

For the running total problem you may want to check this thread http://forum.lessthandot.com/viewtopic.php?f=17&t=7601&st=0&sk=t&sd=a&hilit=running+total%5B/quote%5D

Yes, thanks for the link. There is a good discussion of the other options like ordered updates, SQLCLR, etc.

Interesting to see Moshe Pasumansky mentioned in this article. We studied in the same faculty (school) different departments in my youth (St. Petersburg Technical University).

Cool

/Johan

Viewing 7 posts - 1 through 6 (of 6 total)