Limit of TM1 size

[tryinghard]

Guys,

I have a DB that has over 30 million cells, I am planning to load that in TM1 for reporting. Has anyone dealt with such large number of records. How will be the performance, assume we can get 64 GB Ram on the server and have about 15-20 users accessing data.

[jim wood]

Hi There,

There are quite a few posts on here regarding performance size etc... A couple of pointers for in particular:

1) I my last job we had a cube much bigger than that. It stored around 150k csv lines per week and had 5 years worth of data. We also had around 100 users. It worked fine and ran on a Unix 64-bit box with 64GB of ram.
2) For retrieveing such a large cube you really need to have fast processors. TM1 uses 1 thread for one read. The number of cores helps th number of reads but the speed of those threads will help produce a faster response for each read.

I hope that helps,

Jim.

[tryinghard]

Thank you JIm, so in this cube do you have any rules/feedres at all or is it just loaded through TI processes?

Also, have you ever worked with powerplay, if yes then how would you compare the performance?

[jim wood]

Hello again,

The cube did have rules and feeders. Nothing overly complex but it did have them. As for Power Play I can't say I ever had the pleasure,

Jim.

[lotsaram]

Thank you JIm, so in this cube do you have any rules/feedres at all or is it just loaded through TI processes?

Also, have you ever worked with powerplay, if yes then how would you compare the performance?

Cube build time in TM1 vs. Powerplay is an order of magnitude faster (maybe more, depends on specifics of model). Query time is usually faster or comparable with TM1 unless you have a large rule heavy cube in which case a pre-calculated model is obviously faster. But if you are comparing to powerplay then that's unlikelly to be an issue. TM1's other big advantages are being able to handle much larger cube sizes and much more flexibility with alias and display names and more options in the various UIs.

Where TM1 loses out:
- can be slow browsing BIG dimensions in subset editor
- time dimensions are DIY. To replicate all the nice YTD, last vear, vs LY%, etc time metric functionality that you get with Powerplay OOTB can take quite a lot of work in TM1

I'm sure there are other disadvantages TM1 has but these are the ones that spring immediately to mind.

[stephen waters]

Guys,

I have a DB that has over 30 million cells, I am planning to load that in TM1 for reporting. Has anyone dealt with such large number of records.

Leaving aside data load I query whether 30 million cells is that large. I build demo systems with lots of (often random) data and it is easy to get up to that sort of number with multidimensionality. eg
- 5 years of month\qtr\year = 75
- Account codes and summary lines = 200
- Cost centres and hierarchies = 100
- Versions = 5
- Currency type = 4
Gives 30 million cells.

We have clients with much much larger applications, eg 1500 lines, 2500 cost centres, 10 years data, 40 versions etc etc, many of whom were using TM1 in the old 32 bit days.

I will defer to more experienced contributors over the technical issues when building real life systems but performance will depend on structure of cube and number and quality of calculations. if this is just reporting and you are doing few\simple calcs with lots of consolidation ( and you get feeders etc right) the TM1 performance should be fine. Dont know what overhead the BI front-end adds.

Simplest way is to try it; it is very quick and easy in TM1 to set up a cube (or several different flavours ) load the data and then see what happens!