RAM Issues with 10.2.2 Upgrade

[cipher]

hi All,

I am in a process of upgrading the TM1 10.1 to 10.2.2 FixPack5. We have got a newTest box on which we have 10.2.2 installed. We are planning to bring TM1 Instance running on Prod into newTest Box and see whether we face any issues or not. Below are the specification of servers.

newTest Box:
TM1 10.2.2 FixPack5
Windows 2012R2 Standard Edition
128GB of RAM
2.49GHz, Total 8 Cores

Production Box:
TM1 10.1
Windows 2008R2 Standard Edition
128GB of RAM
2.67GHz, Total 6 Cores

The real problem is here,
The instance which we are planning to migrate from Prod to a newTest Box occupies 72GB of RAM, but after upgrading it to newTest Box, same instance is taking 122GB of RAM and reaches 99% of RAM. I am finding this very strange. I could expect increase by 10% but not this much. Below are the important settings of TM1s.cfg file on newTest Box.

PersistentFeeders=F and T (tried both)
ParallelInteraction=T
MaximumCubeLoadThreads=7
MTQ=ALL

I also checked the Performance monitor, which is Off.

I am really not sure why there is sudden increase in RAM Size. I have done nothing unusual, all I did was, copied data directory from Prod to newTest Box and set up the parameters and turned it's services ON.

Could you guys, please let me know what could be the reason for increase in RAM usage?

Thx!
cipher

[cipher]

Alright Guys, Let me give some more insights...

I turned PersistentFeeder ON, I didnt find any reduction in the RAM Size, but then I restarted the TM1 Instance, and then it came up in very short time and RAM consumption reduced to 57% from 99%. I then loaded one of the yearly file and monitored the RAM consumption but it didnt change. it came to 58%. I then did SaveData and restarted the TM1 Instance and found that there was no spike in RAM consumption, it was around 57%.

So looks like PersistentFeeder=T did work to reduce the RAM consumption of TM1 Instance.

All I need to know is, am I following a right approach by turning PersistentFeeder ON to fix the issue?

[RJ!]

My understanding is that on your first start up, TM1 will do all of the calculations to feed your cubes (and WRITE to your Feeder files), this can lead to a lot of "Garbage" in your RAM. On the 2nd restart, since you have persistent Feeders, TM1 will look for your .FEEDER files and load the cubes based off those, this is why you are seeing the 2nd bounce load much faster & since there are less calculations, there is less "garbage" stored in your RAM.

[gtonkin]

I vaguely remember having an issue like this and traced it back to Parallel Interaction-try searching on the forum for more details but also recollect an IBM document saying that it could increase memory usage by 40%
Set to false to test my memory and yours.

Edit: found the link here

[tomok]

Any time you have MaximumCuveLoadThreads set to more than one there is going to be garbage memory when calculating feeders. The higher the setting the more garbage memory. If .feeders files exist then there is no need to calculate feeders on startup and no garbage memory for them. Your experience makes sense to me.