Migration of If based rules using old rules engine to boolean logic for new rules engine

Mike Potashnick

I am rewriting a model to move from an accelerated process to a new rules engine based rules set.

In the old rules engine where multiple IF conditions could be used, the termination of the IF condition is often either stet) or #NA).

In the new construct is this really necessary or can these just be removed?

an example in the old construct would be:

IF(DB('XXXXXX' !'CLIENT TYPE', 'Income Account to 1')=de.name('Income Account',de.index('Income Account',!'Income Account')),

DB('XXXXXX','Lost Monies Due', !MonthYear, !CLIENTS)

*DB('XXXXXXX',!ENTITY, !MONTHYEAR, !CURRENCY,!CLIENTS, !'CLIENT STATUS', !'CLIENT TYPE', !'FEE TYPE', !LOCATION, !PRODUCT),#NA)

would be replaced with:

(DB('XXXXXXX',!Entity, !'CLIENT TYPE', 'Income Account to 1')=de.name('Income Account',de.index('Income Account',!'Income Account')))*

DB('XXXXXXX','Lost Monies Due', !MonthYear, !CLIENTS)

*DB('XXXXX',!ENTITY, !MONTHYEAR, !CURRENCY,!CLIENTS, !'CLIENT STATUS', !'CLIENT TYPE', !'FEE TYPE', !LOCATION, !PRODUCT)

it seems superfluous to add +#NA or +0*stet as described in the rules engine manual.

Can anyone please provide some guidance?

Thanks

David Swinburne

Hi Mike,

Replacing a rule 

if(a=b,c,#NA) with (a=b)*c is valid but there is one issue to consider.

(a=b) will return either a 1 or a 0, therefore the cube will always be populated with a result when you have a value for c; the IF expression will only populate the cube where the expression is true as a #NA is assigned to the false combinations.

If c has very few values (or a=b is mainly true) then this may not be an issue; it is a function of the data distribution and values.

“it seems superfluous to add +#NA or +0*stet as described in the rules engine manual.”

This is a technique for prefetching data which can greatly improve the performance of an IF block.

Consider an IF block as follows:

IF(a=b,c,d)

The OLAP Server will evaluate this expression and retrieve all data for a and b; it will then retrieve values for c and d based on the result of each tuple. (There are exceptions; in the latest version of the OLAP server I believe c is prefetched by default but this is this basic pattern).

If you have a lot of tuples for a=b then there can be significant latency as the OLAP server retrieves c and d values cell by cell.

By adding +#NA*c*d , e.g.

IF(a=b,c,d) +#NA*c*d

you are forcing the OLAP server to perform a mass retrieval of c and d which can greatly reduce the latency.

An analogy is CellGet vs CellGetC.

There are, however, further exceptions. If d is expensive to calculate and is rarely retrieved based on the values for a and b then prefetching may actually be slower.

As always the Rule Debug log reveals all and helps you with these decisions.

Regards

Dave