TM1 Forum

I have such idea.

Use one TM1 Server for Cubes which contains data for all system. Exchange Rate for example, etc. And other TM1 Servers use for different models - for example first for expenses calculations, secont for HR, third for consolidation. And I supposed to use common data from first TM1 server using Replications

But I see some probelms:
1. I can't replicate cube with picklists. For example I put correspondence for items (is it from PL or BS) using picklist (with elements PL and BS) in the first TM1 server. After it I replicated the cube with correspondence to the second server. But picklist attributes cleared. So I can't change correspondence in the replicated cube in the second server using picklists.

2. I also haven't found ways to run replications using TI process.

So I want to ask. Is it my approach to creating system is normal. Or you can hint me another ways?

If approach isn't bad can you give me advices about tackling that 2 problems?

EP_explorer wrote:1. I can't replicate cube with picklists. For example I put correspondence for items (is it from PL or BS) using picklist (with elements PL and BS) in the first TM1 server. After it I replicated the cube with correspondence to the second server. But picklist attributes cleared. So I can't change correspondence in the replicated cube in the second server using picklists.

Are you replicating the }ElementAttributes_dimname cube?

EP_explorer wrote:2. I also haven't found ways to run replications using TI process.

You can't-- you can schedule it with a chore, though, if that's your goal.

EP_explorer wrote:So I want to ask. Is it my approach to creating system is normal. Or you can hint me another ways?

If approach isn't bad can you give me advices about tackling that 2 problems?

I assume you're using two servers for performance, but in my opinion any performance issues you have can probably be better addressed by a better model design. Replication has always been complex and problematic and, IMHO, is best avoided as much as possible (said as someone who has nine replicating servers).

Matt

Thank you for answers.

Are you replicating the }ElementAttributes_dimname cube?

I tried to do it after your advice but it wasn't very helpful. The thing is that dimensions which are chosen as picklists hasn't transfered to another server. The only way which I see in this case is to replicate cubes with that dimensions

You can't-- you can schedule it with a chore, though, if that's your goal.

It is sad

I assume you're using two servers for performance, but in my opinion any performance issues you have can probably be better addressed by a better model design. Replication has always been complex and problematic and, IMHO, is best avoided as much as possible (said as someone who has nine replicating servers).

My aim is divide TM1 servers for functions (as I have in my models in Cognos Planning where in one library or one apllication I have only that cubes which I need for one purpose - HR planning, or expense planning, or for consolidations and also links between them). So I want to create several TM1 servers which will be divided by functions and change information between them using links. It seems that is quite difficult to maintaing model if I put all cubes (I think it will be more than hundreds) to one TM1 server.

So it seems that replication isn't good way to create links between TM1 servers. How do you advice - is it good idea to use ODBO in TI? or your advice is use one TM1 server?

EP_explorer wrote: My aim is divide TM1 servers for functions (as I have in my models in Cognos Planning where in one library or one apllication I have only that cubes which I need for one purpose - HR planning, or expense planning, or for consolidations and also links between them). So I want to create several TM1 servers which will be divided by functions and change information between them using links.

You can use TM1 Security to hide/show the TM1 cubes.

EP_explorer wrote: It seems that is quite difficult to maintaing model if I put all cubes (I think it will be more than hundreds) to one TM1 server.

more then 100 cubes seems to be quite huge model.... You must be having tens of TM1 developers there and later your client will have to hire a few TM1 Admins to maintain those models.

There could be some cubes which can be merged down into one single cube. I have not seen your Design but there could be things in the Cube design which could be improved.

Do you have same kind of Cubes which are performing the same functions but will be used by different group of people??

You can use TM1 Security to hide/show the TM1 cubes.

Yes, I can. But I meant that I will be difficult to maintain many Cubes in one TM1 Server for administrator

more then 100 cubes seems to be quite huge model.... You must be having tens of TM1 developers there and later your client will have to hire a few TM1 Admins to maintain those models.

There could be some cubes which can be merged down into one single cube. I have not seen your Design but there could be things in the Cube design which could be improved.

Do you have same kind of Cubes which are performing the same functions but will be used by different group of people??

So many Cubes are in Cognos Planning models. I don't think they are very unoptimal. And as I can understand iI will not reduce number of Cubes drastically in TM1. So the question about mantaintanance such system appears.

EP_explorer wrote:So many Cubes are in Cognos Planning models. I don't think they are very unoptimal. And as I can understand iI will not reduce number of Cubes drastically in TM1. So the question about mantaintanance such system appears.

I have been doing TM1 projects for over 15 years, for some of the biggest companies in the U.S., and I've never done an implementation with more then 50 cubes and that was a very unusual case. I can almost guarantee you that a 100 cube system has a lot of redundancies in it. Cognos Planning caused people to create blown out systems because of it's technical limitations. If you are just taking an EP implementation and converting it to TM1, without reviewing and altering the desgn, then you are doing your client a great disservice. Just my opinion, of course.

EP_explorer wrote:

Are you replicating the }ElementAttributes_dimname cube?

I tried to do it after your advice but it wasn't very helpful. The thing is that dimensions which are chosen as picklists hasn't transfered to another server. The only way which I see in this case is to replicate cubes with that dimensions

You could create a dummy cube which contains all the dimensions you need. If you leave the cube blank, it won't take much space, but it will slow replication a little. An alternative would be to replicate the }ElementAttribues_dimname for the dimension you're using in your picklist (this may cause replication to be slower than the earlier option, depending on how often your attributes change as they will be synced each rep). Last, you could use static picklists-- use TI to repopulate the picklist on your star periodically using the dimension as a datasource.

EP_explorer wrote:

I assume you're using two servers for performance, but in my opinion any performance issues you have can probably be better addressed by a better model design. Replication has always been complex and problematic and, IMHO, is best avoided as much as possible (said as someone who has nine replicating servers).

My aim is divide TM1 servers for functions (as I have in my models in Cognos Planning where in one library or one apllication I have only that cubes which I need for one purpose - HR planning, or expense planning, or for consolidations and also links between them). So I want to create several TM1 servers which will be divided by functions and change information between them using links. It seems that is quite difficult to maintaing model if I put all cubes (I think it will be more than hundreds) to one TM1 server.

So it seems that replication isn't good way to create links between TM1 servers. How do you advice - is it good idea to use ODBO in TI? or your advice is use one TM1 server?

If possible, my advice is to use one server. Replication is finicky and more than once my servers have become (silently) desynchronized. The only way to know this has happened is to run manual checks yourself and either refresh the data on the star or re-replicate the desynchronized cubes. Replication performance is pretty poor, but it's magnified by high-latency connections-- since it sounds like you're dealing with two co-located servers this may be less of an issue for you. There's also the issue of maintaining and upgrading two servers instead of one. For me, an upgrade is literally a 1/2 day project since I have to do nine....

Matt

Great thanks for advices. It is really helpful.

I decided to try ODBO connections between TM1 servers. And if it works poorly I will use one TM1 Server although after Cognos Planning it seems really restrictive.

For tomok
I have been doing TM1 projects for over 15 years, for some of the biggest companies in the U.S., and I've never done an implementation with more then 50 cubes and that was a very unusual case. I can almost guarantee you that a 100 cube system has a lot of redundancies in it. Cognos Planning caused people to create blown out systems because of it's technical limitations. If you are just taking an EP implementation and converting it to TM1, without reviewing and altering the desgn, then you are doing your client a great disservice. Just my opinion, of course.

You are absolutely right that Cognos Planning have technical limitations and In many cases TM1 models (which made on Cognos Planning models) are demanded less Cubes. But sometimes it is quite the contrary - you must create addtitional Cubes in TM1 compare with EP models - when you trying to create analog of Accumulation link for example. And also It seems for me that Cognos Planning is more simplier to maintain. Why nobody creates product which contains performance of TM1 and conviniens of Cognos Planning.