Morning,
I put together a very basic set of scaling guidelines :
http://www.observium.org/wiki/Hardware_Scaling
My sample size is quite limited, do any of you have useful numbers from large (>10k ports) installations?
How much RAM does your /ports/ page take? (mouse over the perf on the bottom bar). How much average CPU load does your system have, and what CPU is it? (assuming the host is Observium-only) What storage back-end do you use, and how well is it scaling?
Observium machine (dedicated): 8-core Xeon R5405 @ 2Ghz, 16 GB RAM, 2x 15k 130 GB disk in RAID-1. We are using the poller-wrapper.py script with 16 threads by default.
Devices: 185 Ports: 16668 rrd folder: 29233 MB
/ports/ page: Time: 3.795 sec. Memory: - Cached: 1.86 MB - Page: 96.8 MB - Peak: 97.0 MB
We are using Xcache 3.0.1 PHP opcode cacher as well. From what I can see from the xcache-admin page, we are using approx. 6.23 MB cache.
During polling, CPU usage is approx. 60% on each core. Disk I/O is 615 Ops/sec average...
Any other information you need?