Just to cover all bases, our Orion server (which has NPM, NTA and SAM installed on it) is well spec'd and isn't low on resources. It's a physical server:
2x Intel X5550 2.67Ghz (8 cores each, 16 cores total).
32GB RAM
Server 2008 R2
Network Atlas version: 1.11.0.167
NPM: 10.7
Average CPU load is 10% with 40% memory free. We allocate a minimum of 4GB to SQL and a maximum of 12GB and have enabled Lock Pages in Memory (Enable the Lock Pages in Memory Option (Windows)). SQL performance is excellent. The Orion web interface is snappy too. We have 600 nodes being monitored and a little over 7,000 interfaces.
I've taken a network trace and can't see anything wrong from this point of view.
I've also tried changing the "Net.Tcp Port Sharing Service" config on the Orion server to see if I can get a bit more from the SOAP requests:
- Open Services.
- Look at the properties of the "Net.Tcp Port Sharing Service" service and note the path. e.g. "C:\Windows\Microsoft.NET\Framework64\v4.0.30319".
- Edit the SMSvcHost.exe.config file to up the listenBacklog, maxPendingConnections and maxPendingAccepts. I set these to 100, 1000 and 10 respectively.
- Restart all of the "Net.*" services or restart the server.
However this made no difference at all. Performance of Network Atlas is awful.