Stats by Cube

[Mark RMBC]

I was on a conference call the other day and a TM1 consultant said in passing that the ratio between the "Memory Used for Feeders" and "Number of Fed Cells" ( Memory Used for Feeders/Number of Fed Cells) should ideally be no more than 3:1. I have tried in vain to find a second opinion and can't find anything online to confirm this. If I haven't looked hard enough I can only apologise!

So can anyone please confirm or deny this?

Also, while we are on the subject I have looked online for useful ratios, which include number of "Number of Populated cells" to "Number of Fed Cells" but does a list exist anywhere outlining all the useful ratios?

An example of the statsbycube from one of our models is below:

Memory Used for Views 0
Number of Stored Views 0
Number of Stored Calculated Cells 143976
Number of Populated String Cells 404498
Number of Populated Numeric Cells 3067651
Number of Fed Cells 16599520
Memory Used for Calculations 27397888
Memory Used for Feeders 234848256
Memory Used for Input Data 273755136
Total Memory Used 536001280

cheers, Mark

[declanr]

I have heard other consultants say similar statements; probably with different ratios attached.
I understand the viewpoint of using such a ratio as it gives a starting point for analysis.
However I would say to keep it as just that; every model is different and most cubes within each model are different.

Generalising such things is never as good as having a thorough understanding of what a model is doing and at the end of the day; if the model performs in a quick and acceptable way... who cares what bloody ratio it is.
Horses for courses.

Thats my 2 cents anyway. I am sure there will be plenty of people disagreeing with me to follow...

[lotsaram]

I was on a conference call the other day and a TM1 consultant said in passing that the ratio between the "Memory Used for Feeders" and "Number of Fed Cells" ( Memory Used for Feeders/Number of Fed Cells) should ideally be no more than 3:1.

These aren't even the same unit of measure. This ratio wouldn't just be like comparing apples and oranges but like comparing apples to sombreros. I would guess either you heard wrong or someone was talking out of their ass. 3:1 for this ratio is pretty meaningless

Ratios that would make sense from the available }StatsByCube are
Number of Fed Cells / Number of Populated Numeric Cells
Memory Used for Feeders / Memory Used for Input Data
Memory Used for Input Data / Number of Populated Numeric Cells

The first 2 being measures of the frequency of fed cells to input data (and *might* give some indication of overfeeding), the last one being a measure of storage efficiency in bytes/cell which gives an important indication of whether dimensions are ordered efficiently.

There are "rules of thumb" for the ratio of Number of Fed Cells / Number of Populated Numeric Cells to indicate overfeeding. I have heard some people say 20x but for me personally I wouldn't deem it worthy of investigating unless it was more like 100x. The ratio is only a *presumptive* indication of overfeeding, not proof of overfeeding. I agree 100% with what Declan said; it depends on the context of the model and on the performance. For some modelling there is no alternative but overfeeding. There is no one size fits all for feeding.

[qml]

"Memory Used for Feeders" divided by "Number of Fed Cells" gives you the average memory used by a single feeder flag (in bytes). While I would not consider it to be one of the more important Cube Health Indicator Parameters (CHIPs ), it still does tell you something about the model - specifically how optimal your dimension order is for the type of feeding you have (which, BTW, does not have to be the same as the optimal order for data storage or the optimal order for performance).

Now, since feeders don't need the 8 bytes for data at the end of the trie therefore they take up significantly less memory than data. Maybe a factor of <3 is right, maybe it isn't - honestly I haven't had a model where I would have to care about it. The average size of a data cell is about an order of magnitude more important unless the model is some sort of Olympic Games in feeding.

[Mark RMBC]

Cheers for the responses, much appreciated.

I don't think I heard wrong as I have witnesses! But maybe we all heard wrong.

CHIPS, I will remember that!

Mark

[MarenC]

Hi,

I have checked our model and the grand total for Memory Used for Input Data / Number of Populated Numeric Cells = 112.4

Does this indicate an issue and if so what might that issue be related to?

Just trying to understand what this data is telling me

regards, Maren