Hi All,
Our TM1 application is very slow in loading up. Also when 2 users open cubes simultaneously, they end up with "hanged" sessions. One more issue is that when TI processes run, the cube starts responding slow.
I noticed something - the tm1sd.exe is holding up 19 GB of memory when the total is 21 GB.
deepu_sree wrote:I noticed something - the tm1sd.exe is holding up 19 GB of memory when the total is 21 GB.
This is your key clue as to why the response is slow. You need to go back to your developer and have them do a re-analysis of your system to see whether or not using 19GB makes sense given your data. If that number makes sense (i.e., it's not some sort of design mistake) then you'll need to allocate more RAM to your TM1 service.
deepu_sree wrote:Our TM1 application is very slow in loading up.
As you said your model is about 19Gb in size and loading this large amount of data into the RAM will definatly take some time
Make sure you are using the MaximumCubeLoadThread in your TM1S.CFG file. This will help you to use multiple core while loading the model initially.
deepu_sree wrote:when TI processes run, the cube starts responding slow.
Not sure what your TI Process is doing and what your requirements are. But generally we run the TI Processes over-nigtly (using Chores) so that Users will not get performance issues during the day.
deepu_sree wrote:tm1sd.exe is holding up 19 GB of memory when the total is 21 GB.
As mentioned by tomok, you either have to fine tune your model or add more RAM to your TM1 Server. If possible get some decent RAM size around 35-40Gb.
Folks,
Thanks for your replies..
The total size of our data directory is only 5 GB. Why is that the admin server needs so much space? Our total RAM size is 20 GB.
There is not necessarily a direct relationship between the size on disk of the data directory versus the size in memory of the database when fully loaded. Size in memory will depend on ordering of dimensions in cubes, the amount of rule calculated values and feeders and the size of the calculation cache for each cube. There are already a number of posts on these topics which you should look at.
rkaif wrote:[quote]deepu_sree wrote:Our TM1 application is very slow in loading up.
As you said your model is about 19Gb in size and loading this large amount of data into the RAM will definatly take some time
Make sure you are using the MaximumCubeLoadThread in your TM1S.CFG file. This will help you to use multiple core while loading the model initially.
[/quote]
Please note that using MaximumCubeLoadThread to do a multi-threaded dataload will increase the amount of RAM needed by the server and looking at the situation this will make the server run out of RAM.
Hi,
the reason for the slow load is very simple. You have set many feeder declarations. All of the feeder will be allocated on each server start. Please review the cube rules esp. the feeder definitions. In most of the cases the model is over fed, that means each physical value produces too many feeder marks depending on some imprecise feeder declarations. May be You splash feeders to top sum elements even tough You need one element only.
In 9.5.1 is a nice feature: You can set a new parameter in the tm1s.cfg, PersistentFeeders = T (see reference manual or the thread viewtopic.php?f=3&t=3061). The System should much faster start now, so all of the feeders are stored/cached in a .feeder file which is directly loaded to the memory on startup. Attention! If You decide to use the persistent feeders You may have some funny effects if You change Rules and feeder definitions. You have to switch off the parameter during development on rules.
|手机版|小黑屋|企业绩效管理网
( 京ICP备14007298号 ) 
发表于 2014-3-17 20:06:15
