OpenNMS on a 4-CPU virtual machine with 12GB RAM, with this precise inventory
first router = 1
second level routers that connect to first router = 17
third level switches that connect to second level routers = 82
fourth level end systems of phones, Windows and printers = 400
A 16GB RAM VM is not enough for OpenNMS to discover 1000 simulated
devices. With MIMIC, you can easily test scalability of your NMS
deployment.
An 8GB RAM VM is probably overkill for 1000 nodes. Pay attention to VM storage resource allocation and dedicate resource strictly to OpenNMS. 2-4GB to Java HEAP and the rest to the OS. 1 or 2 cores. Your bottleneck is going to be storage I/O, not RAM or CPU.
ReplyDeleteVerify with JMX collection on the OpenNMS server to watch JVM performance. Setup OpenNMS' native VMWare performance collection and watch host and guest resource statistics to verify whether it is a virtual system starvation issue.
With modern hardware and caching improvements for the collected data store and database performance, disk bottleneck has been pushed much further out, into the 10,000 node range. SAN, SSD, or RAM disks takes it well over 100,000 nodes.
There is a further leap in storage performance coming that scales to millions of nodes, the world of The Internet of Things.