Replication

[appleglaze28]

What is the best architecture for better performance? Cause this is what I'm thinking.

I have a single main server where all data are stored. But I have 2 other machines where users connect. I have 100 hundred users and I want to limit 50 users per machine. This is my assumptions since I'm not too familiar with the replication of TM1. I'll create my a TM1 Server in both machine and replicate the application in my main machine, right? But how can I synchronize the data in both server back to my main server? Can sychronization be scheduled?

How does replication work cause I'm getting this error when I'm trying to replicate the cubes. I'm connected to both the cubes

[Steve Vincent]

Hi appleglaze, i suggest you read the TM1 manuals. There is a whole section on replication and it will answer all those questions for you.

Personally i wouldn't bother limiting users this way. TM1 can cope easily with that number and it would cause more problems than it solved if the only reason to do replication is to split the user load.

[Martin Ingram]

Just posting this for interest of others that may be considering replication - for my current project we originally decided to host centrally and deploy via Citrix;

A large factor in our decision was that we were on a concurrent user model and we would have needed to purchase separate server licenses for each site where we wanted to deploy a server. When we switched to a named user model, we obtained the right to run as many servers as we like (within reason), so long as the total number of users across all servers does not exceed the numbers of named user licenses purchased.

We're now looking at a pilot project to install additional servers in our busiest business units -the only additional cost being the price of the harware, o/s etc.

I did have a chat with a partner yesterday who led me to believe that the license model may have changed yet again from early december, so if in doubt check with your account manager.

[Paul Segal]

Hi Martin,

The balance of opinion seems to be that one central server works better than a number of replicated servers. It'd be useful to know the reasoning behind going towards replicated servers.

Best

[Martin Ingram]

Just a bit of user frustration with Citrix - we'd like to use TM1 for so much more of our offices local reporting needs and Citrix seems to be the main barrier.

We're not definitely going down the replication route, but will pilot it in one location and see how things go

[mattgoff]

We currently have one star and eight planets replicating with it. I can tell you unequivocally that you should stay far, far away from replication. I am in the middle of evaluating my options to get us moved to a single server + Citrix option ASAP. The problems with replication are neverending:

• All cubes on planet are locked during replication, and replication performance over the WAN suffers just as much as do the clients. So, if you need to replicate frequently, be ready for lots of frozen users.
• Logging must be enabled on all replicated cubes as this is how replication is achieved (star transaction log replay on planet and vice versa). If you refresh from external sources frequently, better have lots of log space available.
• There is no integrity checking built-in-- you need to periodically, manually check for errors and, if you find any, delete cube and re-replicate from scratch.
• Control cube replication is dicey. }ElementAttributes seem to work OK (except for aliases on consolidated elements it seems), but }ClientsGroups and }ClientProperties do not (and require jumping through some TI hoops to make work).

Right now I'm dealing with a server that randomly started crashing during replication. After sitting with IBM/Cognos for a week as a P1, I gave up and rebuilt the planet from scratch.

Replication exists in the product, but it's the red-headed stepchild. Anyone that deploys a system using it is asking for trouble. It is not enterprise-grade.

Matt

[Martin Ingram]

Thanks Matt - lots for us to consider (and sorry to hear of your bad experiences).

The way our models are organised, we're actually thinking a TI type approach to replication may be a good option - ie bypass the functionality in the system and use TI to synchronise data and structures on the different boxes (seen older threads discussing this elsewhere on here).

We'll definitely go into this 'test' with our eyes open and i'll gladly post back experiences for others to follow