TM1 10.1 Performance issue

[vichammer]

Hello All.

I uncovered a bug in 10.1 which has been documented as such by IBM. The development is currently working on it. Its related to the new link feature within Performance Modeler. While attempting to use the new feature to create rules for allocations I attempted to use the virtual dimension capability using attributes for a link. The problem is related to use this feature on the source side of the links. While saving and then generating rule creation the application consumed all 24 GB of memory on our server and then went into paging. Unfortunately, since then the server seems to have issues. In particular, running a TI process which loads a cube from a SQL source previously ran int less than 30 seconds and now takes several minutes to complete. Also, it seems that the ViewZeroOut function within the process no longer works. Additionally, a cube which was not a part of any processes and does not have any rules lost all data.

I'm wondering if there is a process to clear out anything within the server that might be causing issues? tp_delete_all?

[lotsaram]

In terms of your original problem it sounds like an automatically generated Performance Modeler allocation rule has had some significant over-feeding problems and blown the memory on your server. The subsequent problems you describe sound a little odd frankly and may or may not be related to your original issue.

The problem with how Performance Modeler creates such rules is that for any missing dimensions the feeder it creates is to the top level "ALL" node of the dimension. If you have several such missing dimensions then you are going to end up with a huge amount of overfeeding which will likely consume significant memory and hamper (if not cripple) performance. The nice pre-sales man in the spiffy suit probably glossed over this along the lines of "we know these feeders are not as efficient as they could be but they are generally pretty good" (at least something along those lines was said at the 10.1 demo I attended). It's all well and good for a demo situation but throw some real data volume or large dimensions at it and there will be problems (as you seem to have encountered.)

Without sounding too trite it may be a good idea to consider getting in an experienced TM1 developer to have a look and sort out the issue.

[vichammer]

Thanks for the feedback lotsaram.

I did get my load process to work normally again. It seems that if I use a public named subset in the cube view for the source it takes the process considerably longer to complete whereas if defaults are selected it seems to run normally. Also, the VeiwZeroOut works as long as I'm not using subsets in the cube view reference within the function. Is this the norm?

[lotsaram]

In terms of the ViewZeroOut not working what is the ViewExtractSkipCalcs setting being used? I suspect that you are assigning a subset to the view that contains only consolidated elements. If you ensure the subset has only leaf elements then the zero out should work fine.