RAM difference between cube and instance

[Catherine]

Hi,

I have an instance where the main cube takes 15 356 Mo in RAM when I have a look at the size in Perspectives Server Explorer. (The following biggest cube takes only 26 Mo in RAM.)
When I have a look at the RAM taken by the instance directly on the server, in the Task Manager or in the Process Explorer, the RAM is 29 041 Mo !
How can you explain the difference ?

I have another instance on the same server which is comparable and the cube RAM is pretty the same as the instance RAM...

Any suggestion would be welcome.
Thanks

[lotsaram]

You also need to consider memory consumption by dimensions. Usually this is not significant but if you have a couple of very large dimensions then this can get up to a few GB. It's also not clear whether the calculation cache and view cache are included in the memory in the cube properties pane.

But I expect the bulk of the difference would be from TM1's internal memory recycling or "garbage memory". If you turn on performance monitor and look in the }StatsForServer cube what does it tell you?

[Catherine]

You're right about the garbage memory.
If I turn on Performance Monitor, I get the following measures:
Memory Used : 29 594 Mo
Memory in Garbage : 13 403 Mo

TM1 operations guide defines Memory in Garbage this way: "TM1 does not release memory back to the operating system while the TM1 server is running. Instead, the server retains unused memory in 'garbage' for later use. Memory that has been allocated to the server but is not in use is represented by this metric."

But could you tell me more about this garbage?
I defined the VMM to 2 Go for my main cube so I think users views are not related to this garbage?
We have some processes running each night after the instance restart. Could that be the views we use in our nightly processes that TM1 stores in the garbage?
Is there a way to limit the memory allocated to this garbage?

Thank you !

[lotsaram]

The largest contributor to garbage memory is redundant processing of feeders at server startup, particularly when using multi-threaded load.

It is only an issue if you don't have the spare RAM capacity, otherwise don't worry about it.

[Catherine]

The largest contributor to garbage memory is redundant processing of feeders at server startup, particularly when using multi-threaded load.

At server startup, I have no problem. Memory in Garbage is very small. And I don't use multi-threaded load.
My problems appear when I run my processes during the night. Globally, the steps are :
-unload calculation rules
-clear data
-rebuild dimensions
-load data
-reload calculation rules
-update security
It multiplies by 2 the RAM used by the instance !

It is only an issue if you don't have the spare RAM capacity, otherwise don't worry about it.

Well it is not on my production server where my instances are restarted each night.
But it can be an issue on my test server where instances are restarted only once a week.
Is there a way to limit the memory allocated to this garbage?