Something that I have vaguely known about for years (for example from Mosha’s post here), but somehow never blogged about, is that using the Root() function in MDX calculations is not great for performance. I’m pretty sure that someone once told me that it was intended for use with defining subcubes for SCOPE statements and not inside calculations at all, which maybe why it hasn’t been optimised. Anyway, here’s an example of the problem and how to work around it.

Take the following query on the Adventure Works cube:

[sourcecode language='text'  padlinenumbers='true' highlight='4']
WITH
MEMBER MEASURES.DEMO AS
([Measures].[Internet Sales Amount],
ROOT([Date]))

SELECT 
{[Measures].[Internet Sales Amount],
MEASURES.DEMO} 
ON 0,
NON EMPTY
[Date].[Calendar].[Calendar Year].MEMBERS
*
[Customer].[Customer].[Customer].MEMBERS
ON 1
FROM
[Adventure Works]
[/sourcecode]

It returns sales for all customers by year, and the calculated measure returns the sales for each customer across all dates using the Root() function.

On a warm SE cache (which means the amount of time taken by the query will be dependent on how quickly SSAS can do the calculation and evaluate the Non Empty) on my laptop this takes a touch over 7 seconds:

Notice also that the Calculate Non Empty End event tells us that the Non Empty filter alone took 2.8 seconds (see here for some more detail on this event).

Now if you rewrite the query, replacing the Root() function with the All Member on the Calendar hierarchy like so:

[sourcecode language='text'  highlight='4']
WITH
MEMBER MEASURES.DEMO AS
([Measures].[Internet Sales Amount],
[Date].[Calendar].[All Periods])

SELECT 
{[Measures].[Internet Sales Amount],
MEASURES.DEMO} 
ON 0,
NON EMPTY
[Date].[Calendar].[Calendar Year].MEMBERS
*
[Customer].[Customer].[Customer].MEMBERS
ON 1
FROM
[Adventure Works]
[/sourcecode]

The query returns the same results, but in just over 5.5 seconds and with the Non Empty taking about 10% of the time it previously took.

I’m making a big assumption here though: the Root() function in the first query returns a tuple containing every All Member from every hierarchy on the Date dimension, not just the All Member from the Calendar hierarchy, so while these two queries return the same results the calculations are not equivalent. You can still get a performance improvement, though, by replacing the Root() function with the tuple it returns, although the resulting MDX will look very messy.

First, to find what the Root() function returns just use a query like this:

[sourcecode language='text' ]
WITH
MEMBER MEASURES.ROOTRETURNS as 
TUPLETOSTR(ROOT([Date]))
SELECT {MEASURES.ROOTRETURNS} ON 0
FROM
[Adventure Works]
[/sourcecode]

Run it in SQL Server Management Studio and you can copy/paste the tuple from the query results:

Here’s the tuple I get from my (somewhat hacked around) Date dimension:

[sourcecode language='text' ]
([Date].[Fiscal].[All Periods],[Date].[Calendar].[All Periods],[Date].[Calendar Weeks].[All Periods],
[Date].[Fiscal Weeks].[All Periods],[Date].[Fiscal Year].[All Periods],[Date].[Date].[All Periods],
[Date].[Calendar Quarter].[All Periods],[Date].[Fiscal Quarter].[All Periods],
[Date].[Calendar Semester].[All Periods],[Date].[Fiscal Semester].[All Periods],
[Date].[Day of Week].[All Periods],[Date].[Day Name].[All Periods],
[Date].[Day of Month].[All Periods],[Date].[Day of Year].[All Periods],
[Date].[Calendar Week].[All Periods],[Date].[Month Name].[All Periods],
[Date].[Calendar Year].[All Periods],[Date].[Fiscal Semester of Year].[All Periods],
[Date].[Calendar Semester of Year].[All Periods],[Date].[Fiscal Quarter of Year].[All Periods],
[Date].[Calendar Quarter of Year].[All Periods],[Date].[Month of Year].[All Periods],
[Date].[Fiscal Week].[All Periods],[Date].[Calendar Week of Year].[All Periods],
[Date].[Fiscal Week of Year].[All Periods],[Date].[Current Date].[All Periods],
[Date].[Is2002].[All Periods],[Date].[Month Day].[All Periods])
[/sourcecode]

Yuck. Anyway, with this gigantic tuple inserted into our calculation like so:

[sourcecode language='text'  highlight='4']
WITH
MEMBER MEASURES.DEMO AS
([Measures].[Internet Sales Amount],
[Date].[Fiscal].[All Periods],[Date].[Calendar].[All Periods],[Date].[Calendar Weeks].[All Periods],[Date].[Fiscal Weeks].[All Periods],[Date].[Fiscal Year].[All Periods],[Date].[Date].[All Periods],[Date].[Calendar Quarter].[All Periods],[Date].[Fiscal Quarter].[All Periods],[Date].[Calendar Semester].[All Periods],[Date].[Fiscal Semester].[All Periods],[Date].[Day of Week].[All Periods],[Date].[Day Name].[All Periods],[Date].[Day of Month].[All Periods],[Date].[Day of Year].[All Periods],[Date].[Calendar Week].[All Periods],[Date].[Month Name].[All Periods],[Date].[Calendar Year].[All Periods],[Date].[Fiscal Semester of Year].[All Periods],[Date].[Calendar Semester of Year].[All Periods],[Date].[Fiscal Quarter of Year].[All Periods],[Date].[Calendar Quarter of Year].[All Periods],[Date].[Month of Year].[All Periods],[Date].[Fiscal Week].[All Periods],[Date].[Calendar Week of Year].[All Periods],[Date].[Fiscal Week of Year].[All Periods],[Date].[Current Date].[All Periods],[Date].[Is2002].[All Periods],[Date].[Month Day].[All Periods])

SELECT 
{[Measures].[Internet Sales Amount],
MEASURES.DEMO} 
ON 0,
NON EMPTY
[Date].[Calendar].[Calendar Year].MEMBERS
*
[Customer].[Customer].[Customer].MEMBERS
ON 1
FROM
[Adventure Works]
[/sourcecode]

The query is a little slower – just over 6 seconds- but still faster than the first query using Root(), and the Non Empty filter is still fast:

Something else to watch out for with Root(), and another good reason not to use it in calculations, is that it returns an error in certain multiselect scenarios as Richard Lees describes here.

In the first blog post in this series a few weeks ago I mentioned that calling Excel and VBA functions from MDX came with a query performance penalty. In this post I’ll give you an illustration of this using the VBA function that I suspect is most frequently called in MDX: the RGB() function.

Take the following MDX query as a baseline:

It returns Countries on columns and all non empty combinations of Date and Product on rows, and the calculated measure returns the value of the Internet Sales Amount measure:

On a SE engine cache it runs in 2.5 seconds on my laptop. With a BACK_COLOR property added to the calculated measure that uses the RGB() function to return the code for red if the measure value is greater than $5000, query performance is a lot worse: it goes up to 6.5 seconds on a warm SE cache.

That’s a big increase just to do some cell highlighting! However in this case the RGB() function can only return two possible integer values, so if you replace the RGB() function with the integers it returns, like so:

…then the query returns in around 3.5 seconds. The last thing to remember is that IIF() statements can perform better if one branch returns null, and in this case we can replace the integer value 16777215 that gives the white background with a null and get the same result:

Now the query returns in around 3 seconds, only 0.5 seconds slower than the original with no colour coding.

Here’s a SSAS Multidimensional MDX tip that I picked up at the PASS Summit back in 2008 at Mosha’s excellent “MDX Deep Dive” precon (incidentally the slides and supporting material are still available here, although a lot of the material is out of date). It’s regarding total-to-date calculations, ie calculations where you are doing a running total from the very first date you have data for up to the current date. The standard way of writing these calculations is something like this:

[sourcecode language='text'  padlinenumbers='true' highlight='3,4,5']
WITH
MEMBER MEASURES.[TTD Sales] AS
SUM(
NULL:[Date].[Calendar].CURRENTMEMBER,
[Measures].[Internet Sales Amount])

SELECT
[Customer].[Country].[Country].MEMBERS 
ON 0,
NON EMPTY
[Date].[Calendar].[Date].MEMBERS
*
[Product].[Product].[Product].MEMBERS
ON 1
FROM
[Adventure Works]
WHERE(MEASURES.[TTD Sales])
[/sourcecode]

This query runs in around 19.2 seconds on my laptop on a cold cache. However if you rewrite it like this:

[sourcecode language='text'  highlight='3,4,5,8,9,10,11,12,13']
WITH
MEMBER MEASURES.[PTTD SALES] AS
SUM(
NULL:[Date].[Calendar].CURRENTMEMBER.PARENT.PREVMEMBER,
[Measures].[Internet Sales Amount])

MEMBER MEASURES.[TTD Sales] AS
MEASURES.[PTTD SALES]
+
SUM(
[Date].[Calendar].CURRENTMEMBER.FIRSTSIBLING:
[Date].[Calendar].CURRENTMEMBER,
[Measures].[Internet Sales Amount])

SELECT
[Customer].[Country].[Country].MEMBERS 
ON 0,
NON EMPTY
[Date].[Calendar].[Date].MEMBERS
*
[Product].[Product].[Product].MEMBERS
ON 1
FROM
[Adventure Works]
WHERE(MEASURES.[TTD Sales])
[/sourcecode]

…it runs slightly faster: around 16.1 seconds on a cold cache on my laptop. Of course this is a very big query, and on most normal queries the difference in performance would be much less significant, but it could still be useful. In fact it’s very similar to the kind of tricks people used to optimise the performance of YTD calculations back in the days of SSAS 2000 – the subject of my second-ever blog post from December 2004! The idea here is that instead of summing up a large set of dates, the calculation sums up all the dates in the current month and then all the months from the beginning of time up to and including the previous full month. For YTD and most other something-to-date calculations trick like this are no longer needed, and indeed are counter-productive and will make your calculations slower. However it seems that for total-to-date calculations they can still help performance.

When you are writing an MDX expression, everywhere you use a set you can give that set a name and then reference the name later on. This is known as creating an inline named set, something I have blogged about a few times (see here and here) over the years. When you are iterating over a set using a function like Generate() or Filter(), if you give that set a name you can then use the Current and CurrentOrdinal functions to find out more about the item in the set returned at the current iteration.

Consider the following MDX query on the Adventure Works cube:

It returns a set of four tuples on rows: every combination of Gender and Marital Status:

If you pass the set on rows to the Filter() function and give it a name (for example MySet) you can then use the CurrentOrdinal function to find the 1-based ordinal of the current iteration. This query uses the CurrentOrdinal function to filter the set shown above so only the first and third items in the set are returned:

With an inline named set you can also use the Current function to return the tuple at the current iteration. Here’s another query that uses the Current function to remove the tuple (Female, Single) from the set:

I won’t pretend that these functions are massively useful, but fans of super-complex MDX will enjoy this vintage post where I used them.

In my last two blog posts (see here and here) I showed how to use the Calculation Evaluation and Calculation Evaluation Detailed Information trace events to work out which MDX calculations are evaluated when a query runs in Analysis Services Multidimensional. That’s very useful, but wouldn’t it be great if you could work out how long any single calculation contributes to the overall duration of a query? If you could, it would make performance tuning MDX calculations much easier.

While you can’t get an exact amount of time taken for each calculation, the good news is that it is possible to get a duration rounded to the next second if your calculation is evaluated in bulk mode.

Take a look at the following query:

[sourcecode language='text'  padlinenumbers='true']
WITH

MEMBER MEASURES.DAYRANK AS
RANK(
[Date].[Date].CURRENTMEMBER, 
[Date].[Date].[Date].MEMBERS)-1

MEMBER MEASURES.HADSALE AS
IIF(
[Measures].[Internet Sales Amount]=0,
NULL,
MEASURES.DAYRANK)

MEMBER MEASURES.LASTSALERANK AS
MAX(
NULL:[Date].[Date].CURRENTMEMBER, 
MEASURES.HADSALE)

MEMBER MEASURES.LASTSALE AS
([Measures].[Internet Sales Amount], 
[Date].[Date].[Date].MEMBERS.ITEM(MEASURES.LASTSALERANK))

MEMBER MEASURES.SIMPLECALC AS
[Measures].[Internet Sales Amount] * 2

SELECT 
HEAD([Customer].[Customer].[Customer].MEMBERS, 200)
*
{MEASURES.SIMPLECALC, MEASURES.LASTSALE}
ON 0,
[Date].[Date].[Date].MEMBERS
ON 1
FROM
[Adventure Works]
[/sourcecode]

This query contains five calculated measures: the first four in the WITH clause, DAYRANK, HADSALE, LASTSALERANK and LASTSALE, are based on my approach for finding the last ever non-empty value for a measure across time; the final measure, SIMPLECALC, is as the name suggests a very simple calculation. On my laptop this query takes around 13 seconds to run on my laptop, on a warm Storage Engine cache. Why does it take so long? It’s clearly the calculations that are the problem, but which one(s)?

Luckily all of the calculations in this query are evaluated in bulk mode so, as I discussed in my last two posts, there is an event raised with:

Event Class = Calculation Evaluation Detailed Information

Event Subclass = 107 – RunEvalNode Finished Calculating Item

…for each of the calculations when they are evaluated. Unfortunately the Duration column for this event always shows 0, but there is a way to see approximately how long the calculation took by comparing the Start Time and Current Time columns in the trace.

The 107 – RunEvalNode event for the measure SIMPLECALC shows the same time for the Start Time and Current Time columns:

This indicates that the SIMPLECALC calculation is evaluated in under a second.

However, sequence of 107 – RunEvalNode events for the LASTSALE calculation shows something different:

There’s a gap of 7 seconds between the StartTime and the CurrentTime, and this indicates that the calculation took 7 seconds to evaluate. It’s a bit frustrating that there isn’t a way to get a more accurate duration here, but it’s still very clear which calculation is taking all the time. Even though the time for calculating LASTSALE includes the time taken for calculating LASTSALERANK, HADSALE and DAYRANK (all of which need to be calculated in order to calculation LASTSALE), the equivalent rows in the trace for these other calculations show they took under a second each. It’s only the logic inside LASTSALE itself that is slow – so that’s where any tuning needs to take place. Indeed, modifying the query to return LASTSALERANK instead of LASTSALE makes the query faster by around 6 seconds, supporting this conclusion.

If you’re curious about what the other 6 seconds of the query execution time is taken up by, it seems like it’s serialisation of the results – something I blogged about here. The query returns a cellset with 400*1190=476000 cells in, and SSAS doesn’t cope well with queries that return a large amount of data.