On internal Atlassian instances we have 16GB on physicaly servers and 8GB inside a virtual machine (KVM).
Base OS is RHEL 5 though migrating to Scientific Linux 6.1. We use Postgres 8.4 on a separate host for the DB.
We try to get at least 16 CPU cores and around 600GB of disk space. All shared artifacts, results and logs are stored on the Bamboo server (after the build), not on the agents. So you want to plan for that.
This scales up pretty well to around 200-400 build plans, 60-120 remote agents. We really try to avoid monolithic, centralised instances choosing instead to give individual Product teams the flexibility and autonomy of their own instance. It has nothing to do with performance, though, it just gives the teams more control and motivation to really own the 'greenness' of their builds.
Bamboo is very chatty so also make sure you have good network connectivity, gigabit ethernet at least. Check the throughput against your source code repos. If you're doing builds based on polling, you'll be hitting them hard.
You'll also need lots of open file descriptors on the server (8192 is a good number but that depends on the number of agents you'll have and how many builds will be executing at any one time)
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.