At Danfoss we are implementing Teamcenter UA globally and need to do the right choice of hardware platform.
In the past we had good experience with running TC Enterprise for 1500 users on HPE Superdome (SD) hardware where we had 2 SD servers running UNIX and Serviceguard. Having the database and the TCEnt application implemented in two separate Serviceguard application packages allows easily to move the database and application individually among the 2 servers that are located closely and connected using state of the art fiber LAN. But I noticed that when DB and application run on the same server, performance is better.
An earlier analysis of database transactions showed a huge overhead of database queries. A simple query for a Part used ~1300 database queries just to return one single item.
These 2 facts leads to my question - what is the significance of the network between database and application in terms of performance? Latency of the LAN is 0 ms but there still must be an overhead in converting to/from TCP-IP protocol and passing 2 network cards compared to communicating between 2 services on the same server. The delay might seem minor but as one query becomes a chain of a huge number of database transactions, the accumulated delay might become considerable.
For the new TCUA environment we are discussing Superdome vs. a number of Windows servers where all communication has to pass the LAN.
Is LAN 0 ms latency a factor in the choice of hardware solution?
Yes, Teamcenter Enterprise and early Teamcenter Unified were "chatty". The latest versions of TcUA has made significant progress reducing the number of round trips required to retrieve data from the database.
The database benefits from RAID 10 while the Enterprise Application servers enjoy RAID 5. Hence, separating the tiers is advised.
With 1500 users, you'll likely want multiple Server Managers (separate servers). Best guess is 3-4. And you'll want multiple Web Application Servers which requires a load balancer (two for redundancy). Then you have Dispatcher, Active Workspace and other elements that could add to your server landscape.
The largest factor in determining hardware is the expected uptime. A high uptime will result in a lot of expensive hardware with some of it sitting idle in case of disaster.
Another factor to consider adding with this many users is WAN acceleration. Cisco WAS is sufficient for Tc while Riverbed is overkill unless you're using it for other applications too.
I've successfully deployed Teamcenter to 1800 users on Windows servers across LAN/WAN.
I'm not sure what you mean by Teamcenter instances. Are these separate databases? Separate DB's require the multisite feature in order to sync between them which incurs some additional overhead and processes to keep ownership clear. Then you would need to consolidate them for your OnePLM solution.
Teamcenter application and web tiers scale very well. Scaling the database can be challenging but isn't really necessary (avoid it - too expensive). Its the integrations across that scale that can be sticky and will cost you the most resources.