I work for a financial institution. We monitor network devices like routers, switches, firewalls and servers with NPM and related products. We operate and unlimited license and currently have around 1200 to 1500 network devices. When servers are added it increases to around 3700 devices or so.
Originally we divided Orion between our two data centers with an instance in each data center, dividing the devices between the data centers around regional lines. Right now the node count is around one third polled out of one data center and two thirds out of the other. In addition to the main pollers we have two in one data center and three in the other. The former instance has around 9400 assigned objects and the latter 18000 assigned objects.
Originally we set it up to spread the polling around and not to over burden Orion with two primary instances as well as managing the polling bandwidth. However, we are interested in possibly using one primary instance with multiple pollers out of one center or the other.
I'm interested in how other comparably sized groups have arranged their polling engines and how they perform. Also what the requirements are in terms of server resources and are you using virtual or physical servers.