About conditional feeders...

[rodrigo]

Hi,

My model have many rules, like 500, I use if's (for example if forecast control is "actual", use "X", if forecasted use "y"), moreless every rule have that kind of situation. My cube grows to 100MB and lasts like 2-3 minutes to do the calc and aggregation work with 7 dimensions.

My performance will grow with conditional feeders as well i can obtain the same accurate results? probably in a pure BI model you can have less formulas to reevaluate each time feeders grow, but in budgets with many formulas reading each conditional feeder is a good idea?

I am using this conditional feeders in some rules that involve many calculations, as well used the reevaluate flag on the .cfg. But before making my move to all the rules i would like have some opinion.

Thanks in advance...

[Eric]

I am not sure I completly understand your question but....

My performance will grow with conditional feeders as well i can obtain the same accurate results?

You will see better performance with conditional feeders and the outcome of the rule will not change, if it is done correctly.

[rodrigo]

Well, I wanted to know the experience with them and if you can use as many conditional feeders as you need them (in my case, to feed a lot of rules..)

[Steve Rowe]

Rodrigo,
It's difficult to give general advice on conditional feeders. They need to be considered on a case by case basis.

It's possible to make the performance of your system worse using conditional feeders, especially if you have re-evaluate conditional feeder flag on.

Perhaps if you post your rules and some structural information the forum can give you more specific advice.

Cheers,

[Martin Ryan]

You can use as many as you like, but nested if statements have (or had) a limit of 30. But then if you write 30 if statements you're likely to go cross eyed.

Also, while conditional statements use less RAM they take up more time as they need to be evaluated. This will affect your server start up time and how long the rule takes to save.

Martin

[jim wood]

Another thing to consider is that conditional feeders will increaes your model startup time. We have quite a big cube set and only one cube contained conditional feeders. Before I removed them the server instance took 6.5 hours to start up. Now I have removed them it is down to 2.5 hours.

Your model may be small at the moment but it is something to consider going forward as your cubes increase in size,

Jim.

[Eric]

Jim,

I believe that is the second time you mentioned the 6.5 hour cube start-up. I am awe struck every time I see that. Would you mind sharing a few others specs on that cube just to satisfy my curiosity?

Dim Count
Elements Per Dim
Cube Size
RAM on Server
Anything else you think is worth sharing?

[jim wood]

The cube in question was called TAPS. It was 9GB in size. It still has 7 dimensions:

Measure - 276
Version - 44
Brand - 251
Handling - 11
Division - 33
Outlet - 7578
Time - 2563

This cube was central hub for everything else and had feeders for 3 other cubes (These were the variable feeders before I remobved them. For Intake alone there were something in the region of 45 variable feeders. I inherited these.).

We Also have the following cubes:
TAPS Sandbox - 8GB
TAPS Summary - 3GB
TAPS Daily Sales Detail - 4GB
TAPS Reporting - 4GB
Intake Planning - 1GB
Plus some otehr smaller cubes.

I inherited TAPS and my long term plan has always been to replace it with the Summary cube. The problem I have been experiencing with 9.1 SP3 have brought my plans forward. I have removed some of the detail from a variety of cubes but remvonig the variable feeders has slashed 4 hours off our server startup time. It has also lowered memory usage by quite a bit.

Jim.

[rodrigo]

somewhere in the applix forum somebody posted the idea of building a forecast cube ( sorry dont remember the name of this excellent collaborator), thats what i am doing, this cube is complex in formulas, but it only works with 2 years of actuals, and a high detailed bugdet forecast for the next year and the ending of the previous period. the matter is that some formulas work on a rolling forecast basis like sales

I am going to develop this ideas and see what happen on my 32 bit 4 GB server,

Thanks guys, interesting answers!