I am currently seeing dramatic increase in size of the model probably due to the usage of the version. Due to the problem I am trying to solve I require many versions (upto 100). This leads to the model size increasing 10 times as fast as using other normal lists as dimensions. for example my mode has 17Mn cells and size is 1.6GB whereas expected is around 200MB. Also I noticed performance hit due to high number of versions present in the model.
Can anyone confirm this is a plausible explanation of the aforementioned behavior. I suspect this is largely due to internal implementation and underlying structure of the Versions . Thanks
It does all depend on the dimensionality of the modules containing versions. The size increase is not just about the increase in the versions list but the product of all of the dimensions where Versions apply
So taking a module that has 3 lists, time and versions with the following members
List A = 5
List B = 10
List C = 100
Time = 50
Versions = 10
The number of cells would be 5*10*100*50*10 = 2,500,000
Adding 1 more Version, the cell count would be 5*10*100*50*11 = 2,750,000 so the increase is proportional
The Gb size then depends on the formatting of line items:
Boolean – 1 Byte
List, Date, Time – 4 Bytes
Text, Number – 8 Bytes
I hope this helps
Hi @DavidSmith ,
For rough estimates I use 10bytes/cell to arrive at the final size . But in the model I mentioned earlier I am seeing 10times more size consumption. I feel that using both version and time in a module has much higher size penalty than normal dimension resulting in same number of cell.This we also tested multiple times. For me this is concern as I am using 120 versions and 365 days as dimension in some modules.
no, it is purely linear
If possible, can you attach a line item extract before and after, along with the Gb size before and after
The other area to look at is lists - any increase in a list element will also add 500b to the size and any list properties will also account for Gb, although not cells. Large lists can, therefore, distort the calculations
Sure @DavidSmith Let me do some testing and will come back with relevant screenshots. Thanks.