TM1 Sizing

[rodrigosborges]

Hi,

I have a TM1 server that processes a large volume of data.
To give an idea of the scale today the server uses 500 GB of RAM.

Unfortunately the server has only 512 GB of RAM and the hardware does not support an expansion of memory.

The client will need to invest a large sum on a new machine, so we need to ensure that this new machine will meet the current and future needs (3 or 4 years), without being too big, which would be a waste of resources and money.

Sizing TM1 is always a matter of discussion, as the "Oficial "book says it is "a fuzzy science".

So how technically defend a "sizing"? What kind of strategy do you usually use?

I will use the recommendations and worksheet sizing of IBM to perform the calculations, but I think the experience of the users of the forum can also help.

The current version of TM1 is 9.5.2 but we are planning to migrate to version 10.

PS: Sorry for English translated by google.

[tomok]

The sizing document is about as good as you are going to get. What they advise you is that each populated cell is going to use 16 bytes. The hard part is estimating how many populated cells you'll have. It's easy when it is a reporting system and the data is coming from a warehouse, you just count the number of rows. For a planning system, or calculation heavy reporting system, you have to estimate the number of calculated cells. Add this number to the number of live data rows and then multiply by 16 bytes. Then give yourself about a 25% margin over that.

[jim wood]

If you are using 512GB at the moment, at the very least I would look to double it. It gives you plenty of room to expand the current service and add additional services as the need may arise. (But that's just me being on the safe side.)

[jstrygner]

Are you confident that RAM used by server in this case is in fact the one that needs to be used?

1. If dimensions order in your cubes is not set in optimal way, you could try to save significant parts of RAM using re-ordering dimensions feature. Unfortunately to do this you would need... RAM , but if single cubes are not huge (less than 10 GB) maybe you could even perform the task on your current box.

2. I saw models where lots of numbers were duplicated with no good reason for it. Maybe some re-work here would also give decent buffer of free memory.